JP2008066782A - Signal processing apparatus and signal processing program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a signal processing apparatus and a signal processing program that exhibit echo suppressing performance which is robust and stable against variation in real delay amount with a small computational complexity when an echo component of a reception input signal and a transmission input signal are out of synchronism with each other (a synchronization shift). <P>SOLUTION: A delay amount estimating unit (TDE) 123 computes a delay amount D by using the reception input signal output from a reception signal storage unit (RBUFF) 121 and the transmission input signal output from a transmission signal storage unit (SBUFF) 122. Then a delay amount quantization unit (DQ) 124 quantizes the delay amount D. An echo canceler unit (EC) 127 suppresses the echo component from a transmission input signal output from an A/D converter (A/D) 18 by referring to a reception input signal output from a reception signal delay processing unit (RDP) 125 and delayed by the delay amount D, and outputs the signal after the echo suppression to a communication unit (COM) 11. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to a signal processing device and a signal processing program, and relates to a process for improving the quality of an audio signal.

  With the spread of the Internet in recent years, a telephone device using VoIP (Voice over IP) technology using the Internet protocol (IP) has been put into practical use. Such a VoIP technology communication apparatus often implements signal processing such as A / D conversion, D / A conversion, encoding / decoding, and acoustic echo suppression processing by software.

  In a call device realized by such software, an operating system (OS) capable of multitask processing, that is, a multitask OS is often used.

  In a multitasking OS, each signal processing is processed as a separate independent task regardless of whether the OS is a real-time OS or a non-real-time OS. Therefore, the true delay amount of the echo component of the received input signal included in the transmitted input signal differs depending on the call, or the true delay amount changes even within the same call. An inconsistency (synchronization shift) of synchronization of input signals occurs. Due to this synchronization shift, there has been a problem that the quality of the audio signal is deteriorated because the acoustic echo suppression processing cannot exhibit sufficient performance.

  In order to cope with the synchronization deviation between the echo component and the transmission input signal due to such a reception input signal, for example, the cross-correlation value of both signals is calculated, and the time when the cross-correlation value is maximum is calculated, A technique is known in which the time at which the pitch similarity is the maximum around the time is estimated as a delay amount (see, for example, Patent Document 1).

Also, there is a technique for calculating the cross-correlation value of both signals, calculating the time at which the cross-correlation value is maximum, obtaining a histogram for the time information, and estimating the time at which the number of appearances is maximum in the histogram as a delay amount. It is known (for example, refer to Patent Document 2).
Japanese Patent Laying-Open No. 2004-282700 (pages 6 to 7, FIGS. 5A and 5B) JP 2004-297236 A (pages 9 to 10 and FIGS. 6 to 9)

  However, the method disclosed in Patent Document 1 described above requires a large amount of calculation because the similarity of the pitch is calculated after the cross-correlation value is calculated. When the amount of calculation is large, the multitask OS has a problem that a synchronization shift between the echo component due to the received input signal and the transmitted input signal tends to occur more frequently.

  In addition, since the delay amount is estimated using the maximum cross-correlation value and the similarity of the pitch, if a small estimation error occurs, the echo suppression processing may be mislearned or the internal state may be erroneously changed, resulting in an echo suppression effect. There was a problem that could not be obtained.

  Further, in the method disclosed in Patent Document 2, a large amount of calculation is required to obtain the maximum frequency from the histogram. In addition, when a change in the true delay amount occurs in the same call, there is a problem that a delay occurs in estimating the delay amount and the echo suppression performance deteriorates.

  The present invention has been made in order to solve the above-described problems, and is a true delay amount with a small amount of calculation with respect to a synchronization mismatch (synchronization shift) between an echo component of a reception input signal and a transmission input signal. An object of the present invention is to provide a signal processing device and a signal processing program that exhibit robust and stable echo suppression performance against fluctuations in the frequency.

  To achieve the above object, a signal processing apparatus according to the present invention includes a first signal storage means for storing a first input signal and a second input signal including an echo component of the first input signal. Using the second signal storage means for storing the first input signal stored in the first signal storage means and the second input signal stored in the second signal storage means. A delay amount estimating unit configured to estimate a delay amount of an echo component included in the second input signal; and an echo suppressing unit configured to suppress at least an echo component included in the second input signal. Quantizes the delay amount with a predetermined quantization step size, and the echo suppression means echo suppresses the first signal delayed by the quantized delay amount as a reference signal of the echo suppression means. It is characterized by.

  According to the present invention, echo suppression that is stable and robust against fluctuations in the true delay amount with a small amount of calculation against the synchronization mismatch (synchronization shift) between the echo component of the received input signal and the transmitted input signal. It is possible to provide a signal processing device and a signal processing program that exhibit performance.

  Embodiments of a signal processing apparatus and a signal processing program according to the present invention will be described below with reference to the drawings.

(First embodiment)
FIG. 1 is a block diagram showing a configuration of a signal processing apparatus according to the first embodiment of the present invention. This signal processing apparatus includes a communication unit (COM) 11, an echo suppression processing unit (ECP) 12, a D / A converter (D / A) 13, a reception amplifier 14, a speaker 15, a microphone 16, It comprises a transmission amplifier 17 and an A / D converter (A / D) 18.

  FIG. 2 is a block diagram showing a configuration of the echo suppression processing unit (ECP) 12. The echo suppression processing unit (ECP) 12 includes a reception signal storage unit (RBUFF) 121 connected to the communication unit (COM) 11 and a transmission signal storage connected to the A / D converter (A / D) 18. Unit (SBUFF) 122, delay amount estimation unit (TDE) 123, delay amount quantization unit (DQ) 124, received signal delay processing unit (RDP) 125, delay amount change detection unit (DCD) 126, It comprises an A / D converter (A / D) 18 and an echo canceller unit (EC) 127 connected to the communication unit (COM) 11.

  FIG. 3 is a block diagram showing a configuration of the echo canceller unit (EC) 127. The echo canceller (EC) 127 includes an adaptive filter (AF) 127 a connected to the received signal delay processor (RDP) 125, an A / D converter (A / D) 18, and a communication unit (COM) 11. A signal subtraction processing unit 127b connected to the reception signal delay processing unit (RDP) 125 and an adaptive filter control unit (CNT) 127c connected to the delay amount change detection unit (DCD) 126.

  The operation of each part of the signal processing apparatus according to the first embodiment of the present invention configured as described above will be described with reference to FIGS.

A communication unit (COM) 11 receives received data that is an encoded digital signal transmitted from a communication partner, and decodes it in units of a predetermined processing time, that is, every frame (F sample). . However, this sampling frequency is assumed to be f _S [Hz]. Then, the communication unit (COM) 11 receives the decoded digital signal for each frame as the received input signal x [n] = (x [nF−F + 1], x [nF−F + 2],. nF]). Here, n indicates a frame number and is an integer of 1 or more.

  Note that the description using the above equal sign is that the received input signal x [n] for one frame is F [nF−F + 1], in which the samples constituting the frame are decoded. , X [nF−F + 2],..., X [nF]. Constituent elements of a transmission output signal s ′ [n], a transmission input signal z [n], a pseudo echo signal y ′ [n], and a residual signal e [n], which will be described later, are indicated by the same description. However, the number of elements constituting these signals is not limited to F, and the case where the number is not F will be described.

  Further, the communication unit (COM) 11 transmits the transmission output signal s ′ [n] = (s ′ [nF−F + 1], s ′) output from the signal subtraction processing unit 127b in the echo suppression processing unit (ECP) 12. [NF−F + 2],..., S ′ [nF]) are input, encoded, and transmitted as transmission data to the communication partner. As will be described later, the transmission output signal s ′ [n] is an echo suppression processing unit (ECP) obtained by converting an A / D converter (A / D) 18 into a digital signal for each frame (F sample). ) 12 is echo-suppressed by 12.

  The D / A converter (D / A) 13 receives the received input signal x [n] output from the communication unit (COM) 11, converts it into an analog signal, and outputs it.

  The receiving amplifier 14 receives the analog signal output from the D / A converter (D / A) 13 as input, amplifies it, and outputs it.

  The speaker 15 receives the amplified analog signal output from the reception amplifier 14 and outputs it as a signal x (t) to the acoustic space.

  The microphone 16 collects the signal z (t) obtained by acoustically coupling the signal x (t) output from the speaker 15 to the acoustic space as described above and the transmitted voice signal s (t), and outputs an analog signal. Convert to and output.

  The transmission amplifier 17 receives the analog signal output from the microphone 16 as input, amplifies it, and outputs it.

  The A / D converter (A / D) 18 receives the amplified analog signal output from the transmission amplifier 17 and converts it into a digital signal for each frame, and transmits the transmission input signal z [n] = ( z [nF-F + 1], z [nF-F + 2],..., z [nF]).

  The echo suppression processing unit (ECP) 12 includes a transmission input signal z [n] output from the A / D converter (A / D) 18, and a reception input signal x [output from the communication unit (COM) 11. n] as an input, the echo component is suppressed from the transmission input signal z [n], and the signal after the echo suppression is transmitted output signal s ′ [n] = (s ′ [nF−F + 1], s ′ [ nF−F + 2],..., s ′ [nF]).

  Next, with reference to FIG. 2, the operation of each unit constituting the echo suppression processing unit (ECP) 12 will be described.

The reception signal storage unit (RBUFF) 121 receives the reception input signal x [n] for each frame output from the communication unit (COM) 11, and echoes the transmission input signal x [n] and the transmission input signal z. Received input signal x _L [n] = (x [nF−L) for L _D + D _MAX samples set in advance based on an expected value of maximum fluctuation range D _MAX of true delay amount D ^* with [n] _D− D _MAX +1], x [nF−L _D −D _MAX +2],..., X [nF]) are stored and output.

Here, L _D is the number of samples set in advance long enough to estimate the true delay amount D ^* in the delay amount estimation unit (TDE) 123 described later, and an adaptive filter unit (described later) AF) The filter length N of the adaptive filter of 127a (represented by the number of samples) is equal to or greater than (N ≦ L _D ).

Transmission signal storage unit (SBUFF) 122 is, A / D converter (A / D) 18 receives the output has been sending input signal z [n] from, _{L D} samples of the transmission input signal _z L [ n] = (z [nF−L _D +1], z [nF−L _D +2],..., z [nF]) are stored and output.

The delay amount estimation unit (TDE) 123 receives the reception input signal x _L [n] for L _D + D _MAX samples output from the reception signal storage unit (RBUFF) 121 and outputs from the transmission signal storage unit (SBUFF) 122. The transmission input signal z _L [n] corresponding to the L _D samples is input, a sample value D that maximizes the cross-correlation is obtained, and is output as an estimated delay amount. Specifically, first, the cross-correlation value P _xz (n, d) between the reception input signal x _L [n] and the transmission input signal z _L [n] is expressed as 0 ≦ d ≦ D _MAX as shown in the following Expression 1. In this interval, d is shifted by one sample, and a cross-correlation value P _xz (n, d) corresponding to each d is calculated.

Here, the time interval for calculating the cross-correlation value P _xz (n, d), that is, the time interval for operating the delay amount estimation unit (TDE) 123 is preferably equal to or less than the time corresponding to L _D samples. . Next, the delay amount estimation unit (TDE) 123 obtains a sample value D that maximizes the cross-correlation value P _xz (n, d) as shown in the following equation 2, and outputs the sampled value D as the estimated delay amount. Here, the function max {·} represents a function that returns the maximum argument.

The delay amount quantizing unit (DQ) 124 receives the estimated delay amount D output from the delay amount estimating unit (TDE) 123 as input, the quantization step size as M, and the quantized delay amount D _OPT and It calculates and outputs the change amount [Delta] D _OPT delay amount _{D OPT} quantized. Here, the amount of change is a difference from the value when the delay amount estimation unit (TDE) 123 estimates the delay amount D.

  Specifically, the actual echo length set in advance (echo length effective for echo suppression) is set to L, and the quantization step size M is set to M = N−L.

FIG. 4 shows the difference between the maximum value and the minimum value of the delay amount that can be suppressed when the delay of the received input signal referred to by the adaptive filter of the adaptive filter unit (AF) 127a described later is fixed. Since the difference between the maximum value and the minimum value of the delay amount that can be suppressed is NL, by setting the quantization step size to NL, the delay amount that cannot be suppressed even when the true delay amount D ^* varies. It can be lost.

Then, the delay amount quantizing unit (DQ) 124 calculates the quantized delay amount D _OPT as in the following Expression 3. Here, the function floor {·} represents a function for truncating the decimal part.

Alternatively, the quantized delay amount D _OPT may be set to a width that satisfies 1 ≦ M ≦ N−L and captures a variation in the true delay amount D ^* . Similarly to the previous case, if it is N−L or less, the delay amount that cannot be suppressed even when the true delay amount D ^* fluctuates can be eliminated.

Alternatively, pre-difference delay is estimated the cross-correlation maximum D and the true delay D ^* may be set as D _H in using the D _H, it is quantized as shown in Equation 4 _{Alternatively} , the delay amount D _OPT may be calculated. The estimated delay amount D having the maximum cross-correlation is influenced by the time when the impulse response is large and does not necessarily match the true delay amount D ^*. Thus, the mismatch can be eliminated. it can. FIG. 5 shows an example in which the estimated delay amount D that maximizes the cross-correlation does not coincide with the true delay amount D ^* . The impulse response h (t) of the echo path and the estimated delay amount that maximizes the cross-correlation at this time D and true delay amount D ^* are shown together.

In FIG. 6, the stepped function of the estimated delay amount D having the maximum cross-correlation and the quantized delay amount D _OPT calculated by Expression 4 is indicated by a bold line. A chain line or a dotted line is a conventional method that does not quantize the delay amount D (for example, Y. Lu, R. Fowler and L. Thompson, “Enhancing Echo Cancellation via Estimation of Delay” IEEE Trans. Signal Process., vol.53, no.11, pp.4159-4168, Nov. 2005.). Compared to such a conventional method, the quantized delay amount D _OPT used in the echo canceller unit (EC) 127 does not fluctuate even if the estimated delay amount D that maximizes the cross-correlation fluctuates within the quantization step size. Therefore, the influence on the echo canceller (EC) 127 due to the estimated fluctuation of the delay time can be reduced.

Further, the amount of change [Delta] D _OPT from quantized delay amount D _OPT of the difference (the current frame and the quantized delay amount D _OPT in one frame before quantized delay amount D _OPT and the current frame (The absolute value of the difference obtained by subtracting the quantized delay amount D _OPT one frame before).

The received signal delay processing unit (RDP) 125 receives received input signals x _L [n] for L _D + D _MAX samples output from the received signal storage unit (RBUFF) 121, and the delay amount quantizing unit (DQ) 124. The received quantized delay amount D _OPT is input, and the received input signal x _N [n] = (x [nF−D _OPT −F−] for F + N samples delayed by the quantized delay amount D _OPT samples. N + 1], x [nF-D _OPT -FN + 2],..., X [nF-D _OPT ]).

The delay amount change detection unit (DCD) 126 receives the change amount ΔD _OPT output from the delay amount quantization unit (DQ) 124 and determines whether the change amount ΔD _OPT is not 0 or a predetermined positive value. Whether or not the quantized delay amount D _OPT has changed is detected, and control information TDEstate [n], which is information indicating whether or not to initialize, is output for each frame. That is, when it is detected that the quantized delay amount D _OPT has changed, the control information TDEstate [n] is set in a state to be initialized and output. When it is detected that the quantized delay amount D _OPT does not change, the control information TDEstate [n] is set to a state in which it is not initialized and output.

The echo canceller unit (EC) 127 has a transmission input signal z [n] output from the A / D converter (A / D) 18 and a delay output from the reception signal delay processing unit (RDP) 125. The received input signal x _N [n] and the control information TDEstate [n] output from the delay amount change detection unit (DCD) 126 are input, and the echo component is suppressed from the transmitted input signal z [n]. The signal after echo suppression is output as a transmission output signal s ′ [n].

  Next, with reference to FIG. 3, the operation of each unit constituting the echo canceller unit (EC) 127 will be described.

  The adaptive filter unit (AF) 127a has a length N filter coefficient h [k, i] (k = 0, 1,..., F−1) (i = 0, 1,..., N−1). ) Is an adaptive filter composed of a variable transversal filter for each sample k.

The adaptive filter unit (AF) 127a is a delayed received input signal x _N [n] output from the received signal delay processing unit (RDP) 125, and an echo-suppressed residual signal output from the signal subtraction processing unit 127b. And the control information ECstate [k] for each sample k output from the adaptive filter control unit (CNT) 127c (k = 0, 1,..., F−1). When the control information ECstate [k] indicates that the learning is not stopped, the filter coefficient h [k, i] is adaptively learned, and the control information ECstate [k] indicates that the learning is stopped. Does not do adaptive learning.

The adaptive filter unit (AF) 127a performs a pseudo operation by performing a convolution operation on the delayed received input signal x _N [n] output from the received signal delay processing unit (RDP) 125 and the filter coefficient h [k, i]. The echo signal y ′ [n] = (y ′ [nF−F + 1], y ′ [nF−F + 2],..., Y ′ [nF]) is calculated and output.

The adaptive filter unit (AF) 127a uses a fixed or variable step size μ _T [k] (k = 0, 1,..., F−1) for controlling the update width of the filter coefficient h [k, i]. And adaptive learning is performed for each sample k.

  The adaptive filter unit (AF) 127a includes, for example, an LMS (Least-Mean-Square) algorithm, an NLMS (Normalized-Least-Mean-Square) algorithm, a learning identification method, an IPNLMS (Improved-Proportinate-NLMS) algorithm, and an affine projection. You may comprise with the adaptive filter based on linear adaptive algorithms, such as (AP: Affine-Projection) algorithm and a recursive least square (RLS) algorithm.

  The adaptive filter unit (AF) 127a is configured by an adaptive filter based on a nonlinear adaptive algorithm such as a gradient-limited learning identification method (Gradient-limited Normalized-Least-Mean-Square) or an adaptive Volterra filter. May be. In this embodiment, an example of a time domain type adaptive filter is shown, but an adaptive filter used in a subband type (band division type) or frequency domain type may be used.

  The signal subtraction processing unit 127b includes the transmission input signal z [n] output from the A / D converter (A / D) 18 and the pseudo echo signal y ′ [n] output from the adaptive filter unit (AF) 127a. ] Is input, and the echo component is subtracted from the transmission input signal z [n] for each sample to suppress the echo component, and the transmission output signal which is the residual signal after the echo suppression. s ′ [n] is output. That is, s ′ [n] = (z [nF−F + 1] −y ′ [nF−F + 1], z [nF−F + 2] −y ′ [nF−F + 2],..., Z [nF] −y ′. [NF]). The transmission output signal s ′ [n] is also output to the communication unit (COM) 11 in addition to each unit in the echo canceller unit (EC) 127.

The adaptive filter control unit (CNT) 127c includes the delayed reception input signal x _N [n] output from the reception signal delay processing unit (RDP) 125 and the control information output from the delay amount change detection unit (DCD) 126. TDEstate [n] and the transmission output signal s ′ [n], which is a residual signal output from the signal subtraction processing unit 127b, are input, and information indicating whether or not the learning is stopped for each sample k. Control information ECstate [k] (k = 0, 1,..., F−1) is output.

Specifically, the adaptive filter control unit (CNT) 127c is (a power value or peak value, referring to these values and power _{characteristics.)} Power characteristics of the received input signal _x N delayed [n] _P X [ k] (k = 0, 1,..., F−1) and power characteristics P _S [k] (k = 0, 1,..., F−1) of the transmission output signal s ′ [n]. Are calculated for each sample k, and it is determined that the learning is stopped when P _S [k]> λ [k] · P _X [k]. If the above inequality does not hold, it is determined that the learning is not stopped.

  Here, λ [k] (k = 0, 1,..., F−1) is an estimated value of the echo path loss, and the filter coefficient h [k, i] (i = 0, 1,. N-1) is calculated for each sample k that has been adaptively learned, and is a variable that decreases as adaptive learning progresses and increases when adaptive learning is incorrect.

Then, the echo canceller (EC) 127 has a filter coefficient h [k, i] (i = 0, 1,..., N−1), a step size μ _T [k], an echo path loss estimated value λ [ k], control information ECstate [k], power characteristic P _X [k] of the received input signal, and power characteristic P _S [k] of the transmission output signal are set as internal states, and the memory in the echo canceller unit (EC) 127 Hold on. Here, the internal state refers to a set of variable values that change at least according to time, and a description thereof will be omitted.

In addition, when the adaptive filter control unit (CNT) 127c indicates that the control information TDEstate [n] output from the delay amount change detection unit (DCD) 126 is in an initializing state, the echo canceller unit (EC) The internal state of 127, in particular the echo path loss estimate λ [k] and the step size μ _T [k] are initialized at the beginning of the frame n.

Thus, by estimating the echo path loss estimated value λ [k] when the quantized delay amount D _OPT changes, the estimated delay including the influence of the variation of the true delay amount D ^* is initialized. It can be prevented that the learning stop state is erroneously determined when the amount changes, and deterioration of the echo suppression performance can be prevented. Further, by initializing the step size μ _T [k] when the quantized delay amount D _OPT changes, the estimated delay amount including the influence of the variation of the true delay amount D ^* is changed. The speed of adaptive learning can be increased, and deterioration of echo suppression performance can be prevented.

  The echo canceller unit (EC) 127 may be configured without the adaptive filter control unit (CNT) 127c. In this case, the adaptive filter unit (AF) 127a operates when the control information ECstate [k] indicates that the learning is not stopped.

  A processing flow of the signal processing apparatus according to the first embodiment configured as described above will be described with reference to FIGS. FIG. 7 is a flowchart showing a processing flow of the entire signal processing apparatus. FIG. 8 shows a delay amount estimation unit (TDE) 123, a delay amount quantization unit (DQ) 124, and a received signal delay processing unit (RDP) 125. FIG. 9 is a flowchart showing a processing flow in the echo canceller unit (EC) 127. FIG.

  In FIG. 7, when there is an outgoing call or incoming call, the communication unit (COM) 11 performs processing for establishing a communication link, and also performs initial setting processing such as initialization of each parameter and each buffer (step S1001). By establishing a communication link, a two-way call with a communication partner becomes possible. When a two-way call is started, a decoder (not shown) in the communication unit (COM) 11 is decoded frame by frame. Read as received input signal x [n]. Further, the transmission input signal z [n] is read via the microphone 16. In addition, the reception signal storage unit (RBUFF) 121 stores the reception input signal x [n], and the transmission signal storage unit (SBUFF) 122 stores the transmission input signal z [n] (step S1002).

Next, as delay amount calculation processing, the received input signal x _L [n] output from the received signal storage unit (RBUFF) 121 and the transmitted input signal z _L output from the transmitted signal storage unit (SBUFF) 122. From [n], the delay amount estimation unit (TDE) 123, the delay amount quantization unit (DQ) 124, the received signal delay processing unit (RDP) 125, and the delay amount change detection unit (DCD) 126 are based on the cross-correlation. The estimated delay amount D calculated in this way is quantized with the quantization step size M, the quantized delay amount D _OPT and the quantized delay amount D _OPT change amount ΔD _OPT are calculated, and the control information TDEstate [n ] Is output (step S1003).

The received signal delay processing unit (RDP) 125 receives the received input signal x _L [n] and the quantized delay amount D _OPT and receives the received input signal x delayed by the quantized delay amount D _OPT sample. _N [n] is output (step S1004). The echo canceller unit (EC) 127 performs echo canceller processing using the delayed received input signal x _N [n] and transmitted input signal z [n] as inputs (step S1005).

  Then, the processing from step S1002 to step S1005 is performed until the call is finished (step S1006).

The delay amount estimation unit (TDE) 123, delay amount quantization unit (DQ) 124, received signal delay processing unit (RDP) 125, and delay amount change detection unit (DCD) 126 shown in FIG. Is called. First, the delay amount estimation unit (TDE) 123 includes a reception input signal x _L [n] output from the reception signal storage unit (RBUFF) 121 and a transmission input signal output from the transmission signal storage unit (SBUFF) 122. Using z _L [n], d is shifted by one sample in the interval of 0 ≦ d ≦ D _MAX , and a cross-correlation value P _xz (n, d) corresponding to each d is calculated (step S1101).

Next, the delay amount estimation unit (TDE) 123 calculates a sample value that maximizes the cross-correlation value P _xz (n, d) as the estimated delay amount D (step S1102). Then, the delay amount quantization unit (DQ) 124 quantizes the estimated delay amount D with the quantization step size M, and the quantized delay amount D _OPT and the quantized delay amount D _OPT change ΔD _OPT is calculated.

Also, the delay amount change detection unit (DCD) 126 detects whether or not the quantized delay amount D _OPT has changed from the change amount ΔD _OPT and controls information TDEstate [ n] is output (step S1103).

The processing of the echo canceller unit (EC) 127 shown in FIG. 9 is performed as follows. First, the pseudo echo signal y is obtained by convolving the delayed reception input signal x _N [n] and the filter coefficient h [k, i] output from the reception signal delay processing unit (RDP) 125 by the adaptive filter unit (AF) 127a. '[N] is calculated, and the signal subtraction processing unit 127b subtracts the pseudo echo signal y' [n] from the transmission input signal z [n] to calculate the transmission output signal s' [n] (step S1). S1201).

  Next, the adaptive filter control unit (CNT) 127c initializes the internal state of the echo canceller unit (EC) 127 according to the control information TDEstate [n] output from the delay amount change detection unit (DCD) 126 as an adaptive filter control process. In addition, control information ECstate [k], which is information indicating whether or not the learning is stopped, is output (step S1202).

  The adaptive filter unit (AF) 127a performs a process of adaptively learning the filter coefficient h [k, i] while receiving control of the control information ECstate [k] output from the adaptive filter control unit (CNT) 127c ( In step S1203), the echo canceller process ends.

  The transmission output signal s ′ [n] output from the echo canceller (EC) 127 in this way is encoded frame by frame by the encoder in the communication unit (COM) 11 and obtained by this encoding. The data is transmitted to the communication partner as transmission data through the communication unit (COM) 11.

  In the above description, the echo canceller (EC) 127 has been described as operating in a time domain type for each sample. A band division type echo canceller using a band division filter such as a filter bank may be realized.

  When estimating the amount of delay based on the cross-correlation between the incoming and outgoing input signals, it is difficult to estimate the exact amount of delay, and there is an estimated fluctuation in the amount of delay that is erroneously estimated with a small amount of delay. And echo suppression performance may be significantly degraded.

  According to the operation of the signal processing apparatus according to the first embodiment described above, since the delay amount quantization unit (DQ) 124 performs the delay amount quantization process, the delay amount estimation error smaller than the quantization step size is small. If so, it is considered that the delay amount of the echo component included in the received input signal does not change. Therefore, the internal state of the echo suppression process is stabilized, and the influence of the estimated fluctuation of the delay amount on the echo suppression process can be reduced, so that a high-quality signal can be output.

  In addition, the estimation accuracy of the estimated delay amount D calculated by the delay amount estimation unit (TDE) 123 by the delay amount quantization processing of the delay amount quantization unit (DQ) 124 is sufficient in the quantization step M. The amount of calculation can be reduced.

(Second Embodiment)
The signal processing device according to the second embodiment is different from the signal processing device according to the first embodiment in that it does not have the echo suppression processing unit (ECP) 12 but has an echo suppression processing unit (ECP) 22. The other parts are the same. Therefore, the same parts are denoted by the same reference numerals and the description thereof is omitted.

  FIG. 10 is a block diagram showing a configuration of the echo suppression processing unit (ECP) 22 of the signal processing device according to the second embodiment of the present invention. The difference between the echo suppression processing unit (ECP) 22 of the signal processing device according to the second embodiment of the present invention and the echo suppression processing unit (ECP) 12 of the signal processing device according to the first embodiment will be described below. . In addition, the same code | symbol is attached | subjected about the same part as the echo suppression process part (ECP) 12 which concerns on 1st Embodiment, and the description is abbreviate | omitted.

  The echo suppression processing unit (ECP) 22 includes a reception signal storage unit (RBUFF) 221 connected to the communication unit (COM) 11 and a transmission signal storage connected to the A / D converter (A / D) 18. Unit (SBUFF) 122, delay amount estimation unit (TDE) 223, delay amount quantization unit (DQ) 224, received signal delay processing unit (RDP) 225, delay amount change detection unit (DCD) 126, It comprises an A / D converter (A / D) 18 and an echo canceller unit (EC) 227 connected to the communication unit (COM) 11.

  FIG. 11 is a block diagram showing a configuration of the echo canceller unit (EC) 227. The echo canceller (EC) 227 includes a frequency domain transform processor (FT) 227a connected to the received signal delay processor (RDP) 225, a frequency domain adaptive filter unit (FDAF) 227b, and a frequency domain inverse transform process. Unit (IFT) 227c, A / D converter (A / D) 18, connection and communication unit (COM) 11, signal subtraction processing unit 227d, frequency domain conversion processing unit (FT) 227e, delay It comprises a frequency domain adaptive filter control unit (FDCNT) 227f connected to a quantity change detection unit (DCD) 126.

  The operation of each part of the signal processing device according to the second embodiment of the present invention configured as described above will be described with reference to FIGS.

The reception signal storage unit (RBUFF) 221 receives the reception input signal x [n] for each frame output from the communication unit (COM) 11, and echoes the transmission input signal x [n] and the transmission input signal z. Received input signal x _L2 [n] = (x [nF−L) for L _D + D _MAX samples set in advance based on an expected value of maximum fluctuation range D _MAX of true delay amount D ^* with [n] _D− D _MAX +1], x [nF−L _D −D _MAX +2],..., X [nF]) are stored and output.

Here, L _D is a sample of a length sufficient to estimate the true delay amount D ^* in the delay amount estimation unit (TDE) 223 described later, and the block size of the echo canceller unit (EC) 227 described later. It is assumed that B (expressed by the number of samples) or more (B ≦ L _D ).

The delay amount estimation unit (TDE) 223 receives the reception input signal x _L2 [n] for L _D + D _MAX samples output from the reception signal storage unit (RBUFF) 221 and outputs from the transmission signal storage unit (SBUFF) 122. The transmission input signal z _L [n] corresponding to the L _D samples is input, a sample value D that maximizes the cross-correlation is obtained, and is output as an estimated delay amount. Specifically, first received input signal _x L2 [n] and the sending input signal _z L [n] cross-correlation value _P xz (n, d) of the as Equation 5 below in _{L D} sample interval, 0 In the interval of ≦ d ≦ D _MAX , d is shifted by one sample, and a cross-correlation value P _xz (n, d) corresponding to each d is calculated.

Next, a sample value D that maximizes the cross-correlation value P _xz (n, d) is obtained as shown in Equation 6 below, and is output as an estimated delay amount.

The delay amount quantizing unit (DQ) 224 has the estimated delay amount D output from the delay amount estimating unit (TDE) 223 as an input, the quantization step size as M, and the quantized delay amount D _OPT and It calculates and outputs the change amount [Delta] D _OPT delay amount _{D OPT} quantized.

The quantization step size M is set to a width that satisfies the inequality 1 ≦ M ≦ B referring to the block size B (= filter length N) of the echo canceller unit (EC) 227 and captures the variation of the true delay amount D ^*. To do. Alternatively, when the actual echo length (echo length effective for echo suppression) is L, the echo length L is measured and calculated in advance, and the quantization step size M is set as M = B−L.

Then, the delay amount quantizing unit (DQ) 224 calculates the quantized delay amount D _OPT as shown in Equation 7 below.

Further, the amount of change [Delta] D _OPT is calculated as the difference between quantized delay amount D _OPT in one frame before quantized delay amount D _OPT and the current frame.

The received signal delay processing unit (RDP) 225 corresponds to the block size B of the echo canceller unit (EC) 227, and the received input signal x for L _D + D _MAX samples output from the received signal storage unit (RBUFF) 221. _L2 [n] and the quantized delay amount D _OPT output from the delay amount quantizing unit (DQ) 224 are input, and 2B (= 2 × B) delayed by the quantized delay amount D _OPT sample The reception input signal x _B [n] = (x [nF−D _OPT −2B + 1], x [nF−D _OPT −2B + 2],..., X [nF−D _OPT ]) for the sample is output.

The echo canceller (EC) 227 receives the transmission input signal z [n] output from the A / D converter (A / D) 18 and the delayed reception received from the reception signal delay processing unit (RDP) 225. Echo from transmission input signal z [n] based on frequency domain adaptive filter algorithm that uses input signal x _B [n] as input and speeds up block adaptive filter using Overlap-Save Method The component is suppressed, and the signal after the echo suppression is set as a transmission output signal s ′ [n] = (s ′ [nF−F + 1], s ′ [nF−F + 2],..., S ′ [nF]). Output.

  Here, the echo canceller unit (EC) 227 makes the block size B equal to the filter length N of an adaptive filter described later, and the overlap preserving method of the echo canceller unit (EC) 227 is the filter of the echo canceller unit (EC) 227. This is a 50% overlap in which twice the length N is FFT (Fast Fourier Transform) points. That is, B = N, and the number of FFT points when performing FFT conversion in the echo canceller unit (EC) 227 is 2B.

  Next, with reference to FIG. 11, the operation of each unit constituting the echo canceller unit (EC) 227 will be described.

The frequency domain transform processing unit (FT) 227a receives the delayed 2B sample received input signal x _B [n] output from the received signal delay processing unit (RDP) 225, and converts it into the frequency domain by 2B point FFT. The frequency spectrum _XFDAF [n, ω] of the received input signal is calculated and output.

The frequency domain adaptive filter unit (FDAF) 227b is a frequency domain adaptive filter configured by a transversal filter having a variable filter coefficient H _FDAF [n, ω].

Further, the frequency domain adaptive filter unit (FDAF) 227b includes the frequency spectrum X _FDAF [n, ω] of the received input signal output from the frequency domain conversion processing unit (FT) 227a and the frequency domain conversion processing unit (FT) 227e. The frequency spectrum S ′ _FDAF [n, ω] of the transmission output signal that is the residual signal after echo suppression output from, and the control information ECstate [n, FD output from the frequency domain adaptive filter control unit (FDCNT) 227f ω] as an input, and the control information ECstate [n, ω] indicates that the learning is not stopped, the filter coefficient H _FDAF [n, ω] is adaptively learned for each frame n and frequency band ω, and the control information ECstate is obtained. When [n, ω] indicates that the learning is stopped, adaptive learning is not performed. In this way, the filter coefficient H _FDAF [n, ω] is calculated and stored in the frequency domain adaptive filter unit (FDAF) 227b.

Further, the frequency domain adaptive filter unit (FDAF) 227b has a frequency spectrum X _FDAF [n, ω] of the received input signal output from the frequency domain conversion processing unit (FT) 227a and a filter coefficient H _FDAF [n, ω]. _Is used to calculate and output the frequency spectrum Y ′ _FDAF [n, ω] of the pseudo echo signal as Y ′ _FDAF [n, ω] = H _FDAF [n, ω] · X _FDAF [n, ω].

The frequency domain adaptive filter unit (FDAF) _227b performs adaptive learning using a variable step size μ _F [n, ω] that controls the update width of the filter coefficient H _FDAF [n, ω].

  Note that the frequency domain adaptive filter unit (FDAF) 227b may be configured by an adaptive filter based on a linear adaptive algorithm such as an LMS algorithm, an NLMS algorithm, or a sequential least square algorithm.

  Further, the frequency domain adaptive filter unit (FDAF) 227b may be configured with an adaptive filter based on a nonlinear adaptive algorithm such as a gradient-limited learning identification method or an adaptive Volterra filter.

The frequency domain inverse transform processing unit (IFT) 227c receives the frequency spectrum Y ′ _FDAF [n, ω] of the pseudo echo signal output from the frequency domain adaptive filter unit (FDAF) 227b as an input, and a 2B point IFFT (Inverse Fast). Fourier transform is performed, and the final F samples are converted into pseudo echo signals y ′ _FDAF [n] = (y ′ _FDAF [nF−F + 1], y ′ _FDAF [nF−F + 2],..., Y ′ _FDAF [ nF]) and output.

The signal subtraction processing unit 227d transmits the transmission input signal z [n] output from the A / D converter (A / D) 18 and the pseudo echo signal y output from the frequency domain inverse conversion processing unit (IFT) 227c. ' _FDAF [n] is input and the pseudo echo signal y' _FDAF [n] is subtracted for each sample from the transmission input signal z [n] to suppress the echo component and send the signal after the echo suppression. Speech output signal s ′ [n] = (s ′ [nF−F + 1], s ′ [nF−F + 2],..., S ′ [nF]) = (z [nF−F + 1] −y ′ _FDAF [nF −F + 1], z [nF−F + 2] _−y′FDAF [nF−F + 2],..., Z [nF] _−y′FDAF [nF]).

The frequency domain transform processing unit (FT) 227e receives the transmission output signal s ′ [n], which is a residual signal after echo suppression output from the signal subtraction processing unit 227d, and receives 2B−F samples from the beginning. Zero padding is performed, and finally the F sample transmission output signal s ′ [n] is arranged and converted to the frequency domain by performing 2B-point FFT conversion, and the frequency spectrum S ′ of the transmission output signal is converted. _FDAF [n, ω] is calculated and output.

The frequency domain adaptive filter control unit (FDCNT) 227f outputs the frequency spectrum X _FDAF [n, ω] of the received signal output from the frequency domain conversion processing unit (FT) 227a and the frequency domain conversion processing unit (FT) 227e. The frequency spectrum S ′ _FDAF [n, ω] of the transmission output signal, which is the residual signal, and the control information TDEstate [n] output from the delay amount change detection unit (DCD) 126 are input, and the frame n Control information ECstate [n, ω], which is information indicating whether or not the learning is stopped, is output for each frequency band ω.

Specifically, first, the frequency domain adaptive filter control unit (FDCNT) 227f includes the power spectrum of the received signal | X _FDAF [n, ω] | ² and the power spectrum of the transmitted output signal | S ′ _FDAF [n, ω]. | ² is calculated for each frame n and frequency band ω. If the inequality | S ′ _FDAF [n, ω] | ² > λ _FDAF [n, ω] · | X _FDAF [n, ω] | ² holds, it is determined that the learning is stopped. Here, λ _FDAF [n, ω] is an estimated value of the echo path loss, and is a variable amount that decreases when adaptive learning progresses and increases when adaptive learning is incorrect. Also, λ _FDAF [n, ω] is calculated by updating the filter coefficient H _FDAF [n, ω] for each frame n and frequency band ω that are adaptively learned. If the above inequality does not hold, it is determined that the learning is not stopped.

The echo canceller unit (EC) 227 includes a filter coefficient H _FDAF [n, ω], a step size μ _F [n, ω], an echo path loss estimated value λ _FDAF [n, ω], and control information ECstate [n, ω], power spectrum of received signal | X _FDAF [n, ω] | ² , power spectrum of transmitted output signal | S ′ _FDAF [n, ω] | ² as an internal state, echo canceller unit (EC) 227 Held in the memory.

In addition, when the frequency domain adaptive filter control unit (FDCNT) 227f indicates that the control information TDEstate [n] output from the delay amount change detection unit (DCD) 126 is in a state of being initialized, the echo canceller unit ( EC) 227 _initializes the internal state, in particular the echo path loss estimate λ _FDAF [n, ω] and the step size μ _F [n, ω].

In this way, when the quantized delay amount D _OPT changes, the estimated value λ _FDAF [n, ω] of the echo path loss is initialized, so that the estimation including the effect of the variation of the true delay amount D ^* is included. It is possible to prevent the learning stop state from being erroneously determined when the delay amount is changed, and to prevent deterioration of the echo suppression performance. Further, by initializing the step size μ _F [n, ω] when the quantized delay amount D _OPT changes, the estimated delay amount including the influence of the variation of the true delay amount D ^* can be obtained. The speed of adaptive learning can be increased at the time of fluctuation, and deterioration of echo suppression performance can be prevented.

  In this embodiment, the echo canceller (EC) 227 is composed of a frequency domain type adaptive filter that is not gradient constrained. However, the echo canceller unit (EC) 227 is a gradient constrained frequency domain type adaptive filter. It may be configured. The echo canceller (EC) 227 shows an example of the overlap storage method in the present embodiment, but may be configured by an overlap addition method (Overlap-Add Method). The echo canceller (EC) 227 shows an example using FFT conversion in the present embodiment, but MDCT conversion may be used.

  A processing flow of the signal processing apparatus according to the second embodiment configured as described above will be described with reference to FIGS. FIG. 12 is a flowchart showing a processing flow of the entire signal processing apparatus. FIG. 13 shows a delay amount estimation unit (TDE) 223, a delay amount quantization unit (DQ) 224, a received signal delay processing unit (RDP) 225, and a delay. 5 is a flowchart showing a flow of processing in a quantity change detection unit (DCD) 126.

  FIG. 14 is a flowchart showing the flow of processing in the echo canceller (EC) 227. In addition, about the same operation step as the operation | movement of the signal processing apparatus which concerns on 1st Embodiment demonstrated with reference to FIGS. 7-9, the same code | symbol is attached | subjected and description of the part is abbreviate | omitted.

In FIG. 12, after step S1002, the received signal input signal x _L2 [n] output from the received signal storage unit (RBUFF) 221 and the transmitted message output from the transmitted signal signal storage unit (SBUFF) 122 as delay amount calculation processing. From the input signal z _L [n], a delay amount estimation unit (TDE) 223, a delay amount quantization unit (DQ) 224, a received signal delay processing unit (RDP) 225, and a delay amount change detection unit (DCD) 126 are The estimated delay amount D calculated based on the cross correlation is quantized with the quantization step size M, and the quantized delay amount D _OPT and the quantized delay amount D _OPT change amount ΔD _OPT are calculated and controlled. Information TDEstate [n] is output (step S2003).

The received signal delay processing unit (RDP) 225 receives the received input signal x _L2 [n] and the quantized delay amount D _OPT and receives the received input signal x delayed by the quantized delay amount D _OPT sample. _B [n] is output (step S2004). The echo canceller unit (EC) 227 performs echo canceller processing using the delayed received input signal x _B [n] and transmitted input signal z [n] as inputs (step S2005).

  Then, according to the determination processing in step S1006, the processing from step S1002 to step S2005 is performed until the call ends.

The processing of the delay amount estimation unit (TDE) 223, delay amount quantization unit (DQ) 224, received signal delay processing unit (RDP) 225, and delay amount change detection unit (DCD) 126 shown in FIG. 13 is performed as follows. Is called. First, the delay amount estimation unit (TDE) 223 includes a reception input signal x _L2 [n] output from the reception signal storage unit (RBUFF) 221 and a transmission input signal output from the transmission signal storage unit (SBUFF) 122. Using z _L [n], d is shifted by one sample in the interval of 0 ≦ d ≦ D _MAX , and a cross-correlation value P _xz (n, d) corresponding to each d is calculated (step S2101).

Next, the delay amount estimation unit (TDE) 223 calculates the sample value that maximizes the cross-correlation value P _xz (n, d) as the estimated delay amount D (step S2102). Then, the delay amount quantization unit (DQ) 224 quantizes the estimated delay amount D with the quantization step size M, and the quantized delay amount D _OPT and the quantized delay amount D _OPT change amount ΔD _OPT is calculated. Also, the delay amount change detection unit (DCD) 126 detects whether or not the quantized delay amount D _OPT has changed from the change amount ΔD _OPT and controls information TDEstate [ n] is output (step S2103).

The processing of the echo canceller unit (EC) 227 shown in FIG. 14 is performed as follows. First, the frequency domain transform processing unit (FT) 227a converts the delayed received input signal x _B [n] output from the received signal delay processing unit (RDP) 225 into the frequency domain by FFT conversion, and receives the received input signal. Frequency spectrum X _FDAF [n, ω] is calculated.

Next, the frequency domain adaptive filter unit (FDAF) 227b uses the frequency spectrum X _FDAF [n, ω] of the received input signal and the filter coefficient H _FDAF [n, ω] to _generate the frequency spectrum Y ′ _FDAF [ n, ω] is calculated. Then, the frequency domain inverse transform processing unit (IFT) 227c performs IFFT conversion on the frequency spectrum _Y′FDAF [n, ω] of the pseudo echo signal to calculate the pseudo echo signal _y′FDAF [n]. Then, the signal subtraction processing unit 227d subtracts the pseudo echo signal y ′ _FDAF [n] from the transmission input signal z [n] to calculate the transmission output signal s ′ [n] (step S2201).

  Next, the frequency domain transform processing unit (FT) 227e transforms the transmission output signal s ′ [n] into the frequency domain by FFT transform, and the frequency domain adaptive filter control unit (FDCNT) 227f performs delay as the adaptive filter control processing. The internal state of the echo canceller (EC) 227 is initialized according to the control information TDEstate [n] output from the quantity change detector (DCD) 126, and control information ECstate [n, ω] is output (step S2202).

The frequency domain adaptive filter unit (FDAF) 227b receives the control of the control information ECstate [n, ω] output from the frequency domain adaptive filter control unit (FDCNT) 227f, and the filter coefficient H _FDAF [n, ω] The adaptive learning process is performed (step S2203), and the echo canceller process ends.

  According to the operation of the signal processing apparatus according to the second embodiment described above, the delay amount quantization unit (DQ) 224 is similar to the delay amount quantization unit (DQ) 124 of the first embodiment. Since the processing is performed, a high-quality signal can be output. Further, the amount of calculation for estimating the delay amount D of the delay amount estimation unit (TDE) 223 can be reduced.

(Third embodiment)
The signal processing device according to the third embodiment is different from the signal processing device according to the first embodiment in that it does not have the echo suppression processing unit (ECP) 12 but has an echo suppression processing unit (ECP) 32. The other parts are the same. Therefore, the same parts are denoted by the same reference numerals and the description thereof is omitted.

  FIG. 15 is a block diagram showing a configuration of an echo suppression processing unit (ECP) 32 of the signal processing device according to the third embodiment of the present invention. The difference between the echo suppression processing unit (ECP) 32 of the signal processing device according to the third embodiment of the present invention and the echo suppression processing unit (ECP) 12 of the signal processing device according to the first embodiment will be described below. . In addition, the same code | symbol is attached | subjected about the same part as the echo suppression process part (ECP) 12 which concerns on 1st Embodiment, and the description is abbreviate | omitted.

  The echo suppression processing unit (ECP) 32 includes a reception signal storage unit (RBUFF) 321 connected to the communication unit (COM) 11 and a transmission signal storage connected to the A / D converter (A / D) 18. Unit (SBUFF) 122, delay amount estimation unit (TDE) 323, delay amount quantization unit (DQ) 324, received signal delay processing unit (RDP) 325, and A / D converter (A / D) 18 And an echo reduction unit (ER) 327 connected to the communication unit (COM) 11.

  FIG. 16 is a block diagram showing the configuration of the echo reduction unit (ER) 327. The echo reduction unit (ER) 327 includes a frequency domain conversion processing unit (FT) 327a connected to the reception signal delay processing unit (RDP) 325 and a frequency connected to the A / D converter (A / D) 18. Area conversion processing unit (FT) 327b, received power calculation unit (POW) 327c, transmission power calculation unit (POW) 327d, acoustic coupling amount estimation unit (ACLE) 327e, and echo amount estimation unit (ELE) 327f A frequency domain control unit (FCNT) 327g, a gain storage unit (GTBL) 327h, an echo suppression gain calculation unit (GCAL) 327i, a signal suppression unit (SS) 327j, and a communication unit (COM) 11. Frequency domain inverse transform processing unit (IFT) 327L.

  The operation of each part of the signal processing device according to the third embodiment of the present invention configured as described above will be described with reference to FIGS.

The reception signal storage unit (RBUFF) 321 receives the reception input signal x [n] for each frame output from the communication unit (COM) 11, and echoes the transmission input signal x [n] and the transmission input signal z. Received input signal x _L3 [n] = (x [nF−L) for L _D + D _MAX samples set in advance based on an expected value of maximum fluctuation range D _MAX of true delay amount D ^* with [n] _D− D _MAX +1], x [nF−L _D −D _MAX +2],..., X [nF]) are stored and output. Here, L _D is the number of samples long enough to estimate a true delay amount D ^* in a delay amount estimation unit (TDE) 323 described later, and the number of FFT points in the frequency domain transform processing unit (FT) 327a. L and is _F or more _{(L F} ≦ _{L D).}

The delay amount estimation unit (TDE) 323 receives the reception input signal x _L3 [n] for L _D + D _MAX samples output from the reception signal storage unit (RBUFF) 321 and outputs from the transmission signal storage unit (SBUFF) 122. The transmission input signal z _L [n] corresponding to the L _D samples is input, a sample value D that maximizes the cross-correlation is obtained, and is output as an estimated delay amount. Specifically, first as in the received input signal _x L3 [n] and the sending input signal _z L [n] of the cross-correlation value _P xz (n, d) the following formula 8 with the _{L D} sample interval, 0 In the interval of ≦ d ≦ D _MAX , d is shifted by one sample, and a cross-correlation value P _xz (n, d) corresponding to each d is calculated.

Next, the delay amount estimation unit (TDE) 323 obtains a sample value D that maximizes the cross-correlation value P _xz (n, d) as shown in the following Expression 9, and outputs the sampled value D as the estimated delay amount.

The delay amount quantizing unit (DQ) 324 includes the estimated delay amount D output from the delay amount estimating unit (TDE) 323 and the frequency domain transform processing unit (FT) 327a constituting the echo reduction unit (ER) 327. The window function w [i] (i = 0, 1,..., L _F ) is input, the quantization step size is M, and the quantized delay amount D _OPT is calculated and output.

FIG. 17 shows an example of the window function w [i] (i = 0, 1,..., L _F ) used in the frequency domain transform processing unit (FT) 327a and the frequency domain transform processing unit (FT) 327b. . Here the length of the window function w [i] is that except when it is 0, the effective window length L _W. That is, the effective window length L _W is L _W ≦ L _F when compared with the FFT point number L _F of the frequency domain transform processing unit (FT) 327a and the frequency domain transform processing unit (FT) 327b.

The delay amount quantization unit (DQ) 324 satisfies 1 ≦ M ≦ L _W −L _O and true if the quantization step size M is L _{O as the} overlap of the window function w [i]. It is set so as to capture fluctuations in the delay amount D ^* . Then, the quantized delay amount D _OPT is calculated as in Expression 10 below.

In this way, by setting the quantization step size M with reference to the effective window length L _W and overlap L _O of the window function w [i], the effective window length L _W can be calculated from the FFT point number L _F. Even when the asymmetric window becomes smaller or when the window function w [i] dynamically changes with time, the influence of the estimated fluctuation of the delay amount on the echo suppression processing can be reduced.

The received signal delay processing unit (RDP) 325 includes received input signals x _L3 [n] for L _D + D _MAX samples output from the received signal storage unit (RBUFF) 321, and the delay amount quantizing unit (DQ) 324. Received input signal x _W [n] = (x [nF−D _OPT −F + 1], x with F samples delayed by the corrected delay amount D _OPT samples, using the output corrected delay amount D _OPT as input. [NF−D _OPT −F + 2],..., X [nF−D _OPT ]) are output.

  Next, with reference to FIG. 16, the operation of each unit constituting the echo reduction unit (ER) 327 will be described.

The echo reduction unit (ER) 327 receives the delayed reception input signal x _W [n] output from the reception signal delay processing unit (RDP) 325 and the transmission output from the A / D converter (A / D) 18. The speech input signal z [n] is input, the echo component is suppressed from the transmission input signal z [n], and the signal after the echo suppression is transmitted as the transmission output signal s ′ [n] = (s ′ [nF− F + 1], s ′ [nF−F + 2],..., S ′ [nF]) for each frame. Further, the window function w [i] is output.

The frequency domain transform processing unit (FT) 327a receives the delayed received input signal x _W [n] output from the received signal delay processing unit (RDP) 325, and receives the frequency spectrum X [n, ω] of the received input signal. Is calculated and output. Specifically, using the F samples of the received input signal x _{W [n],} using combined frame past received input signal x _{W [n-1],} also, L _F samples of the signal, such as by zero padding generates, this window function w [i] (i = 0,1 , ···, L F) to windowing and converted to the frequency domain by the FFT transformation of _{L F} point, the received input signal The frequency spectrum X [n, ω] is calculated and output.

Here, Expressions 11-1, 11-2, and 11-3 below show an example of the window function w [i] when F = 160 and L _F = 512. At this time, effective window lengths are L _W = 2 × F = 320, and overlap L _O = F = 160.

  The frequency domain transform processing unit (FT) 327a outputs a window function w [i] used by the frequency domain transform processing unit (FT) 327a.

The frequency domain transform processing unit (FT) 327b receives the transmission input signal z [n] output from the A / D converter (A / D) 18, and uses the frequency spectrum Z [n, ω of the transmission input signal. ] Is calculated and output. Specifically, using the F samples of the transmission input signal z [n], using combined frame past of the transmission input signal z [n-1], also, L _F samples of the signal, such as by zero padding generates, this window function w [i] (i = 0,1 , ···, L F) to windowing and converted to the frequency domain by the FFT transformation of _{L F} point, sending the input signal Frequency spectrum Z [n, ω] is calculated and output.

The received power calculation unit (POW) 327c receives the frequency spectrum X [n, ω] of the received input signal output from the frequency domain transform processing unit (FT) 327a, and receives the received power spectrum | X [ n, ω] | ² is calculated and output. Since the acoustic coupling amount does not usually change abruptly in time, it is possible to estimate the acoustic coupling amount with higher accuracy by using a smoothed value than by using an instantaneous value. Therefore, the received power calculation unit (POW) 327c For example, as shown in Expression 12, the received power spectrum | X _S [n, ω] | ² smoothed using the value | X _S [n−1, ω] | ² of the previous frame is calculated and output. However, α _X [ω] is preferably about 0.75 to 0.999.

The transmission power calculation unit (POW) 327d receives the frequency spectrum Z [n, ω] of the transmission input signal output from the frequency domain conversion processing unit (FT) 327b as an input, and the transmission power spectrum that is the power spectrum thereof. | Z [n, ω] | ² is calculated and output. Since the amount of acoustic coupling usually does not change suddenly in time, the amount of acoustic coupling can be estimated more accurately by using a smoothed value than by using an instantaneous value, so the transmission power calculation unit (POW) 327d is For example, as shown in Equation 13, the transmission power spectrum | Z _S [n, ω] | ² smoothed using the value | Z _S [n−1, ω] | ² of the previous frame is calculated and output. To do. However, α _Z [ω] is preferably about 0.75 to 0.999.

The acoustic coupling amount estimation unit (ACLE) 327e includes a smoothed reception power spectrum | X _S [n, ω] | ² output from the reception power calculation unit (POW) 327c, and a transmission power calculation unit (POW) 327d. The smoothed transmission power spectrum | Z _S [n, ω] | ² output from the control information ERstate [n, ω] output from the frequency domain control unit (FCNT) 327 g and the frequency band For each ω, the acoustic coupling amount | H [n, ω] | ² is calculated by, for example, the following Expression 14.

Then, the acoustic coupling amount estimation unit (ACLE) 327e calculates and outputs a smoothed acoustic coupling amount | H _S [n, ω] | ² using a value one frame before as in the following Expression 15. However, α _H [ω] is preferably about 0.03 to 0.99.

Here, at the beginning of the call, for example, by increasing α _H [ω] for about 5 seconds from the start of the call, the update of the acoustic coupling amount | H _S [n, ω] | ² is accelerated. By doing so, since the acoustic coupling amount is initialized at the beginning of the call, it is possible to prevent the suppression amount from being reduced at the beginning of the call.

However, when the control information ERstate [n, ω] output from the frequency domain control unit (FCNT) 327g indicates that the learning is stopped, the acoustic coupling amount estimation unit (ACLE) 327e does not update the acoustic coupling amount. Then, the past acoustic coupling amount | H _S [n−1, ω] | ² is used.

The echo amount estimation unit (ELE) 327f includes the smoothed reception power spectrum | X _S [n, ω] | ² output from the reception power calculation unit (POW) 327c and the acoustic coupling amount estimation unit (ACLE) 327e. With the output acoustic coupling amount | H _S [n, ω] | ² as an input, the echo amount | Y [n, ω] | ² included in the frequency spectrum Z [n, ω] of the transmission input signal is As shown in Equation 16, the frequency band ω is estimated.

The echo amount estimation unit (ELE) 327f uses the smoothed value to make the signal after echo suppression more natural than using the instantaneous echo amount | Y [n, ω] | ^2. Then, the echo amount | Y _S [n, ω] | ² that is smoothed using the value of the previous frame as in the following Expression 17 is calculated and output for each frequency band ω. However, α _Y [ω] is preferably about 0.7 to 0.99.

The frequency domain control unit (FCNT) 327g receives the smoothed reception power spectrum | X _S [n, ω] | ² output from the reception power calculation unit (POW) 327c and the acoustic coupling amount estimation unit (ACLE) 327e. acoustic coupling of one frame before output _{| H S [n-1,} ω] | ² and as input, control information ERstate per frame n and frequency bands omega is information indicating whether a learning stopped [N, ω] is output.

Specifically, when the acoustic coupling amount changes abruptly, that is, when | H _S [n, ω] | ² > β _H [ω] · | H _S [n−1, ω] | ² is satisfied. If the received input signal is not sufficiently large, that is, if | X _S [n, ω] | ² <β _X [ω] is satisfied, the frequency domain control unit (FCNT) 327 g It is determined that However, β _H [ω] is preferably about 0.9 to 30. β _X [ω] is desirably about 30 dB to 40 dB. On the other hand, when it is not in any of the above cases, it is determined that the learning is not stopped.

  The gain storage unit (GTBL) 327h stores and outputs a parameter γ [ω] that controls a preset nonlinear echo suppression amount. However, γ [ω] is preferably about 1.0 to 2.0.

The echo suppression gain calculation unit (GCAL) 327i includes a smoothed transmission power spectrum | Z _S [n, ω] | ² output from the transmission power calculation unit (POW) 327d, and an echo amount estimation unit (ELE). The smoothed echo amount | Y _S [n, ω] | ² output from the 327 f and the parameter γ [ω] output from the gain storage unit (GTBL) 327 h are input, and the echo suppression gain G [n, ω ] Is calculated and output.

Specifically, the echo suppression gain calculation unit (GCAL) 327i calculates the echo suppression gain G [n, ω] by Equation 18 using the Wiener Filter method.

  Also, the echo suppression gain calculation unit (GCAL) 327i prevents the transmission voice quality from deteriorating due to excessive echo suppression, and prevents the background noise from being intermittently suppressed. ], A background noise level is calculated for each frequency band using a signal level of a section that is not a voiced section, and the echo suppression gain G [n, ω] is controlled so as not to be suppressed below the background noise level. I do not care.

The signal suppression unit (SS) 327j outputs the frequency spectrum Z [n, ω] of the transmission input signal output from the frequency domain transform processing unit (FT) 327b and the echo suppression gain calculation unit (GCAL) 327i. Using the echo suppression gain G [n, ω] as an input, the echo of the frequency spectrum Z [n, ω] of the transmission input signal is suppressed, and the spectrum S ′ [ n, ω].

  The frequency domain inverse transform processing unit (IFT) 327L receives the frequency spectrum S ′ [n, ω] of the transmission output signal output from the signal suppression unit (SS) 327j as an input, and transmits the transmission output signal s ′ [n]. = (S ′ [nF−F + 1], s ′ [nF−F + 2],..., S ′ [nF]) is calculated and output to the communication unit (COM) 11.

Specifically, the frequency domain inverse transform unit (IFT) 327 L, the frequency spectrum S '[n, ω] of the transmitter output signal is converted into a time region by IFFT conversion of _{L F} point, frequency domain transform processor Corresponding to the windowing by the window function w [i] (i = 0, 1,..., L _F ) of the part (FT) 327b, the overlap L _O samples are appropriately transmitted output signals of past frames. A process of returning the overlap is performed by superimposing s ′ [n] on the transmission output signal s ′ [n] of the current frame, and the transmission output signal s ′ [n] = (s ′ [nF−F + 1], s ′ [nF−F + 2],..., s ′ [nF]) are calculated and output.

  A processing flow of the signal processing apparatus according to the third embodiment configured as described above will be described with reference to FIGS. FIG. 18 is a flowchart showing a processing flow of the entire signal processing apparatus. FIG. 19 shows a delay amount estimation unit (TDE) 323, a delay amount quantization unit (DQ) 324, and a received signal delay processing unit (RDP) 325. FIG. 20 is a flowchart showing the flow of processing in the echo reduction unit (ER) 327. In addition, about the same operation step as the operation | movement of the signal processing apparatus which concerns on 1st Embodiment demonstrated with reference to FIGS. 7-9, the same code | symbol is attached | subjected and description of the part is abbreviate | omitted.

In FIG. 18, after step S <b> 1002, as a delay amount calculation process, the received input signal x _L3 [n] output from the received signal storage unit (RBUFF) 321 and the transmission output from the transmitted signal signal storage unit (SBUFF) 122. From the speech input signal z _L [n], the delay amount estimation unit (TDE) 323, the delay amount quantization unit (DQ) 324, and the received signal delay processing unit (RDP) 325 are estimated based on the cross-correlation. The delay amount D is quantized with the quantization step size M, and the quantized delay amount D _OPT is calculated and output (step S3003).

The received signal delay processing unit (RDP) 325 receives the received input signal x _L3 [n] and the quantized delay amount D _OPT and receives the received input signal x delayed by the quantized delay amount D _OPT sample. _W [n] is output (step S3004). The echo reduction unit (ER) 327 performs echo reduction processing using the delayed reception input signal x _W [n] and transmission input signal z [n] as inputs (step S3005).

  Then, according to the determination processing in step S1006, the processing from step S3002 to step S3005 is performed until the call ends.

The processing of the delay amount estimation unit (TDE) 323, the delay amount quantization unit (DQ) 324, and the reception signal delay processing unit (RDP) 325 shown in FIG. 19 is performed as follows. First, the delay amount estimation unit (TDE) 323 receives the reception input signal x _L3 [n] output from the reception signal storage unit (RBUFF) 321 and the transmission input signal output from the transmission signal storage unit (SBUFF) 122. Using z _L [n], d is shifted by one sample in the interval of 0 ≦ d ≦ D _MAX , and a cross-correlation value P _xz (n, d) corresponding to each d is calculated (step S3101).

Next, the delay amount estimation unit (TDE) 323 calculates the sample value that maximizes the cross-correlation value P _xz (n, d) as the estimated delay amount D (step S3102). Then, the delay amount quantizing unit (DQ) 324 quantizes the estimated delay amount D with the quantization step size M, and calculates the quantized delay amount D _OPT (step S3103).

The processing of the echo reduction unit (ER) 327 shown in FIG. 20 is performed as follows. First, the frequency domain transform processing unit (FT) 327a and the frequency domain transform processing unit (FT) 327b use the received input signal x _W [n] and the transmitted input signal z [n] for F samples, respectively, for one frame. using together past signal, also generates the L _F samples of the signal by zero padding, the receiving frame is a buffer for converting into the frequency domain, and updates the transmission frame (step S3201r, S3201s).

Next, frequency domain transform processor (FT) 327a is windowed window function w [i] in L _F samples of the signal stored in the received frame, transformed to the frequency domain by the FFT transformation of L _F point Then, the frequency spectrum X [n, ω] of the received input signal is calculated (step S3202r). Then, the received power calculation unit (POW) 327c calculates the received power spectrum | X [n, ω] | ² and the smoothed received power spectrum | X _S [n, ω] | ² (step S3203r).

Similarly, frequency domain transform processor (FT) 327b is windowed window function w [i] in L _F samples of the signal stored in the transmission frame, in a frequency domain by the FFT transformation of L _F point conversion to the frequency spectrum Z [n, omega] of the transmission input signal is calculated (steps S3202s), sending power calculating unit (POW) 327d is sending power spectrum ^{| Z [n, ω] |} 2 and The smoothed transmission power spectrum | Z _S [n, ω] | ² is calculated (step S3203s).

Then, the frequency domain controller (FCNT) 327 g outputs a control information ERstate [n, ω], acoustic coupling amount estimating unit (ACLE) 327e is smoothed received power spectrum _{| X S [n, ω]} | 2 And the smoothed transmission power spectrum | Z _S [n, ω] | ² and the control information ERstate [n, ω] as inputs, calculate the acoustic coupling amount | H _S [n, ω] | ² (step S3204).

Echo amount estimating unit (ELE) 327f is acoustic coupling amount | contained in ² and transmission input signal as an input ^{_{| H s [n, ω]}} | 2 and smoothing the received power spectrum _{| X} S [n, _ω] The echo amount | Y _S [n, ω] | ² is estimated (step S3205).

Next, the echo suppression gain calculation unit (GCAL) 327i includes the smoothed transmission power spectrum | Z _S [n, ω] | ² output from the transmission power calculation unit (POW) 327d, and the echo amount estimation unit. (ELE) The smoothed echo amount | Y _S [n, ω] | ² output from the 327 f and the parameter γ [ω] output from the gain storage unit (GTBL) 327 h are input, and the echo suppression gain G [ n, ω] is calculated (step S3206).

  The signal suppression unit (SS) 327j receives the frequency spectrum Z [n, ω] of the transmission input signal and the echo suppression gain G [n, ω] calculated by the echo suppression gain calculation unit (GCAL) 327i. Then, the echo component of the frequency spectrum Z [n, ω] of the transmission input signal is suppressed (step S3207).

  Finally, the frequency domain inverse transform processing unit (IFT) 327L performs frequency inverse transform processing on the frequency spectrum S ′ [n, ω] of the transmission output signal output from the signal suppression unit (SS) 327j, and overlaps. , The transmission output signal s ′ [n] is calculated (step S3208), and the echo reduction process ends.

  In the above description, the echo reduction unit (ER) 327 has been described as operating as a method of processing for each frequency band by frequency domain conversion by FFT conversion. You may implement | achieve the frequency domain type | mold echo suppression processing, such as the method which puts together the frequency band by FFT conversion, and processes for every frequency band group, the method using MDCT conversion, and band division filters, such as a filter bank.

  According to the operation of the signal processing apparatus according to the third embodiment described above, the delay amount quantization unit (DQ) 324 is similar to the delay amount quantization unit (DQ) 124 of the first embodiment. Since the processing is performed, a high-quality signal can be output. Further, the amount of calculation for estimating the delay amount D of the delay amount estimation unit (TDE) 323 can be reduced.

Furthermore, the amount of delay quantization processing delay amount quantizer (DQ) 324, with reference to the effective window length L _W of the window function w [i] by setting the quantization step size M, valid window If the length L _W is small becomes asymmetrical windows than FFT points L _F and, even when the window function w [i] is changed dynamically by time, reduce the effects of estimation fluctuation of the delay amount provided to the echo suppression process can do.

(Fourth embodiment)
The signal processing device according to the fourth embodiment is different from the signal processing device according to the first embodiment in that it does not have the echo suppression processing unit (ECP) 12 but has an echo suppression processing unit (ECP) 42. The other parts are the same. Therefore, the same parts are denoted by the same reference numerals and the description thereof is omitted.

  FIG. 21 is a block diagram showing a configuration of an echo suppression processing unit (ECP) 42 of the signal processing device according to the fourth embodiment of the present invention. The echo suppression processing unit (ECP) 42 includes a reception signal storage unit (RBUFF) 421, a reception signal delay processing unit (RDP) 4251, a reception signal delay processing unit (RDP) 4252, and a signal selection unit (SEL) 428. And a first echo canceller unit (EC1) 4271 and a second echo canceller unit (EC2) 4272.

  The first echo canceller (EC1) 4271 includes a frequency domain transform processor (FT) 4271a, a frequency domain adaptive filter unit (FDAF) 4271b, a frequency domain inverse transform processor (IFT) 4271c, and a signal subtraction. A processing unit 4271d, a frequency domain conversion processing unit (FT) 4271e, and a frequency domain adaptive filter control unit (FDCNT) 4271f are included.

  Similarly, the second echo canceller unit (EC2) 4272 includes a frequency domain transform processing unit (FT) 4272a, a frequency domain adaptive filter unit (FDAF) 4272b, a frequency domain inverse transform processing unit (IFT) 4272c, It comprises a subtraction processing unit 4272d, a frequency domain conversion processing unit (FT) 4272e, and a frequency domain adaptive filter control unit (FDCNT) 4272f.

When the actual echo length (echo length effective for echo suppression) is L and the block sizes of the first echo canceller (EC1) 4271 and the second echo canceller (EC2) 4272 are both B, 1 ≦ M time satisfying _D ≦ B-L to set the shift width _{M D} in advance. In this embodiment, there are two echo canceller units, a first echo canceller unit (EC1) 4271 and a second echo canceller unit (EC2) 4272, but the number q of echo canceller units and the received input signal x [n ] Is set so that the relationship of D _MAX ≦ q · M _D is established in the maximum fluctuation width D _MAX of the true delay amount D ^* between the echo and the transmission input signal z [n]. The number q of echo cancellers is set according to the value of _D.

  The operation of each part of the echo suppression processing unit (ECP) 42 of the signal processing device according to the fourth embodiment of the present invention configured as described above will be described with reference to FIG.

The received signal storage unit (RBUFF) 421 receives the received input signal x [n] for each frame output from the communication unit (COM) 11, and receives the first echo canceler unit (EC1) 4271 and the second echo. with reference to the block size B of the canceller unit _{(EC2) 4272, 2B + M} D samples of the received input signal _{x B0 [n] = (x} [nF-2B-M D +1], x [nF-2B-M D +2 ],..., X [nF]) are stored and output.

Received signal delay processing unit (RDP) 4251 is received signal storage unit (RBUFF) the received input signal of 2B + _{M D} samples outputted from the 421 _x B0 [n] as input, received input signal 2B samples _{x B1} [ n] = (x [nF-2B + 1], x [nF-2B + 2],..., x [nF]) is output.

Received signal delay processing unit (RDP) 4252 is to the received signal storage unit (RBUFF) 421 output from the 2B + _{M D} samples of the received input signal _x B0 [n] and an _{input, x} B0 [n] of the past from among Receive input signal x _B2 [n] = (x [nF−2B−M _D +1], x [nF−2B−M _D +2],..., X [nF−M _D for 2B samples corresponding to time ]) Is output. That, and outputs the received input signal corresponding to the _{M D} samples past than received signal delay processor (RDP) received input signal output from the 4251 _x B1 [n] _x B2 as [n].

The first echo canceler unit (EC1) 4271 is transmitted from the transmission input signal z [n] output from the A / D converter (A / D) 18 and output from the reception signal delay processing unit (RDP) 4251. Based on a frequency domain adaptive filter algorithm in which the received input signal x _B1 [n] is input and the block adaptive filter is speeded up using the overlap preserving method, the echo component is suppressed from the transmitted input signal z [n], The signal after the echo suppression is the residual signal e ₁ [n] = (e ₁ [nF−F + 1], e ₁ [nF−F + 2],..., E ₁ [nF]) = (z [nF−F + 1). ] _-Y'FDAF [nF-F + 1], z [nF-F + 2] _-y'FDAF [nF-F + 2],..., Z [nF] _-y'FDAF [nF]).

The second echo canceller (EC2) 4272 outputs the transmission input signal z [n] output from the A / D converter (A / D) 18 and the reception signal delay processing unit (RDP) 4252. The received incoming signal x _B2 [n] is used as an input, and an echo component is suppressed from the transmitted input signal z [n] based on a frequency domain adaptive filter algorithm that speeds up the block adaptive filter using the overlap preserving method. Then, the signal after the echo suppression is output as a residual signal e ₂ [n] = (e ₂ [nF−F + 1], e ₂ [nF−F + 2],..., E ₂ [nF]).

  Here, the first echo canceller unit (EC1) 4271 and the second echo canceller unit (EC2) 4272 make the block size B equal to the filter length N, and the first echo canceller unit (EC1) 4271 and the second echo canceller unit (EC1) 4271 The echo preserving part (EC2) 4272 of the overlap preservation method is 50% overlap in which twice the filter length N is used as the FFT score. That is, B = N and the number of FFT points is 2B.

The frequency domain transform processing unit (FT) 4271a receives the reception input signal x _B1 [n] for 2B samples output from the reception signal delay processing unit (RDP) 4251 and converts it into the frequency domain by 2B point FFT. The frequency spectrum X ₁ [n, ω] of the received input signal is calculated and output.

Similarly, the frequency domain transform processing unit (FT) 4272a receives the reception input signal x _B2 [n] corresponding to 2B samples output from the reception signal delay processing unit (RDP) 4252, and enters the frequency domain by the 2B point FFT. After conversion, the frequency spectrum X ₂ [n, ω] of the received input signal is calculated and output.

The frequency domain adaptive filter unit (FDAF) 4271b is a frequency domain adaptive filter configured by a transversal filter having a variable filter coefficient H ₁ [n, ω]. Then, the frequency spectrum X ₁ [n, ω] of the received input signal output from the frequency domain transform processing unit (FT) 4271a and the residual after echo suppression output from the frequency domain transform processing unit (FT) 4271e. The control information EC1state [n, ω] is learned by using the frequency spectrum E ₁ [n, ω] of the signal and the control information EC1state [n, ω] output from the frequency domain adaptive filter control unit (FDCNT) 4271f as inputs. The filter coefficient H ₁ [n, ω] is adaptively learned for each frame n and the frequency band ω when indicating that it is not in the stopped state, and is adaptive when the control information EC1state [n, ω] indicates that it is in the learning stopped state. Do not learn. In this way, the filter coefficient H ₁ [n, ω] is calculated and output.

Further, the frequency domain adaptive filter unit (FDAF) 4271b has a frequency spectrum X ₁ [n, ω] of the received input signal output from the frequency domain transform processing unit (FT) 4271a and a filter coefficient H ₁ [n, ω]. Are used to calculate and output the frequency spectrum Y ′ ₁ [n, ω] of the pseudo echo signal as Y ′ ₁ [n, ω] = H ₁ [n, ω] · X ₁ [n, ω].

The frequency domain adaptive filter unit (FDAF) 4271b performs adaptive learning using a variable step size μ _F1 [n, ω] that controls the update width of the filter coefficient H ₁ [n, ω].

Similarly, the frequency domain adaptive filter unit (FDAF) 4272b is a frequency domain adaptive filter composed of a transversal filter with a variable filter coefficient H ₂ [n, ω]. Then, the frequency spectrum X ₂ [n, ω] of the received input signal output from the frequency domain transform processing unit (FT) 4272a and the echo-residual residual signal output from the frequency domain transform processing unit (FT) 4272e. the frequency spectrum _E 2 [n, ω] and control information EC2state [n, ω] output from the frequency domain adaptive filter control section (FDCNT) 4272f as input and control information EC2state [n, ω] is the learning stop When it indicates that the state is not a state, the filter coefficient H ₂ [n, ω] is adaptively learned for each frame n and frequency band ω, and when the control information EC2state [n, ω] indicates that the learning is stopped, adaptive learning is performed. Do not do. In this way, the filter coefficient H ₂ [n, ω] is calculated and output.

Further, the frequency domain adaptive filter unit (FDAF) 4272b has a frequency spectrum X ₂ [n, ω] of the received input signal output from the frequency domain transform processing unit (FT) 4272a and a filter coefficient H ₂ [n, ω]. Are used to calculate and output the frequency spectrum Y ′ ₂ [n, ω] of the pseudo echo signal as Y ′ ₂ [n, ω] = H ₂ [n, ω] · X ₂ [n, ω].

The frequency domain adaptive filter unit (FDAF) 4272b performs adaptive learning using a variable step size μ _F2 [n, ω] that controls the update width of the filter coefficient H ₂ [n, ω].

  The frequency domain adaptive filter unit (FDAF) 4271b and the frequency domain adaptive filter unit (FDAF) 4272b include, for example, an adaptive filter based on a linear adaptive algorithm such as an LMS algorithm, an NLMS algorithm, and a sequential least square algorithm, a gradient-limited learning identification method, an adaptation It consists of adaptive filters based on non-linear adaptive algorithms such as Volterra filters.

The frequency domain inverse transform processing unit (IFT) 4271c receives the frequency spectrum Y ′ ₁ [n, ω] of the pseudo echo signal output from the frequency domain adaptive filter unit (FDAF) 4271b as input and performs 2B-point IFFT transform. The final F samples are calculated as pseudo echo signals y ′ ₁ [n] = (y ′ ₁ [nF−F + 1], y ′ ₁ [nF−F + 2],..., Y ′ ₁ [nF]). Output.

Similarly, the frequency domain inverse transform processing unit (IFT) 4272c receives the frequency spectrum Y ′ ₂ [n, ω] of the pseudo echo signal output from the frequency domain adaptive filter unit (FDAF) 4272b as an input, and a 2B point IFFT. Conversion is performed, and the final F samples are converted into a pseudo echo signal y ′ ₂ [n] = (y ′ ₂ [nF−F + 1], y ′ ₂ [nF−F + 2],..., Y ′ ₂ [nF]). Is calculated and output.

The signal subtraction processing unit 4271d receives the transmission input signal z [n] output from the A / D converter (A / D) 18 and the pseudo echo signal y output from the frequency domain inverse conversion processing unit (IFT) 4271c. ' ₁ [n] as an input, the pseudo echo signal y' ₁ [n] is subtracted from the transmission input signal z [n] for each sample to suppress the echo component, and the residual signal after the echo suppression E ₁ [n] = (e ₁ [nF−F + 1], e ₁ [nF−F + 2],..., E ₁ [nF]) = (z [nF−F + 1] −y ′ ₁ [nF−F + 1] ], Z [nF−F + 2] −y ′ ₁ [nF−F + 2],..., Z [nF] −y ′ ₁ [nF]).

Similarly, the signal subtraction processing unit 4272d transmits the transmission input signal z [n] output from the A / D converter (A / D) 18 and the pseudo signal output from the frequency domain inverse conversion processing unit (IFT) 4272c. as input and the echo signal y _'2 [n], the residual signal after echo suppression _{e 2 [n] = (e} 2 [nF-F + 1], e 2 [nF-F + 2], ···, e 2 [NF]) = (z [nF−F + 1] −y ′ ₂ [nF−F + 1], z [nF−F + 2] −y ′ ₂ [nF−F + 2],..., Z [nF] −y ′ ₂ [NF]) is output.

The frequency domain transform processing unit (FT) 4271e receives the echo-suppressed residual signal e ₁ [n] output from the signal subtraction processing unit 4271d, performs zero padding of 2B-F samples from the beginning, and finally The residual signal e ₁ [n] for F samples is placed in the frequency domain by performing the 2B point FFT transform to calculate the frequency spectrum E ₁ [n, ω] of the transmission output signal. And output.

Similarly, the frequency domain transform processing unit (FT) 4272e receives the echo-suppressed residual signal e ₂ [n] output from the signal subtraction processing unit 4272d, and performs zero padding of 2B-F samples from the beginning. Finally, e ₂ [n] for F samples is arranged and converted to the frequency domain by performing 2B-point FFT conversion to calculate the frequency spectrum E ₂ [n, ω] of the transmission output signal And output.

The frequency domain adaptive filter control unit (FDCNT) 4271f outputs the frequency spectrum X ₁ [n, ω] of the received signal output from the frequency domain conversion processing unit (FT) 4271a and the frequency domain conversion processing unit (FT) 4271e. The control signal EC1state [n, ω], which is information indicating whether or not the learning is stopped, is output for each frame n and frequency band ω, with the frequency spectrum E ₁ [n, ω] of the residual signal as an input. To do.

Specifically, first, the frequency domain adaptive filter control unit (FDCNT) 4271f has a power spectrum | X ₁ [n, ω] | ² of the received signal and a power spectrum | E ₁ [n, ω] | ^{2 of the} residual signal. Are calculated for each frame n and frequency band ω. When the inequality | E ₁ [n, ω] | ² > λ ₁ [n, ω] · | X ₁ [n, ω] | ² holds, it is determined that the learning is stopped. On the other hand, if this inequality does not hold, it is determined that the learning is not stopped.

Here, λ ₁ [n, ω] is an estimated value of the echo path loss, and is a variable amount that decreases as adaptive learning progresses and increases when adaptive learning is incorrect. Also, λ ₁ [n, ω] is calculated by updating the filter coefficient H ₁ [n, ω] for each adaptively learned frame n and frequency band ω.

Similarly, the frequency domain adaptive filter control unit (FDCNT) 4272f includes the frequency spectrum X ₂ [n, ω] of the received signal output from the frequency domain conversion processing unit (FT) 4272a and the frequency domain conversion processing unit (FT). The frequency spectrum E ₂ [n, ω] of the residual signal output from the 4272e is input, and control information EC2state [n, ω which is information indicating whether or not the learning is stopped for each frame n and the frequency band ω. ] Is output.

Specifically, first, the frequency domain adaptive filter control unit (FDCNT) 4272f includes the power spectrum | X ₂ [n, ω] | ² of the received signal and the power spectrum | E ₂ [n, ω] | ^{2 of the} residual signal. Are calculated for each frame n and frequency band ω. If the inequality | E ₂ [n, ω] | ² > λ ₂ [n, ω] · | X ₂ [n, ω] | ² holds, it is determined that the learning is stopped. On the other hand, if this inequality does not hold, it is determined that the learning is not stopped.

Here, λ ₂ [n, ω] is an estimated value of the echo path loss, and is a variable amount that decreases as adaptive learning progresses and increases when adaptive learning is incorrect. Also, λ ₂ [n, ω] is calculated by updating the filter coefficient H ₂ [n, ω] for each frame n and frequency band ω that are adaptively learned.

  The first echo canceller (EC1) 4271 and the second echo canceller (EC2) 4272 show examples of frequency domain type adaptive filters without gradient constraints in this embodiment. A type adaptive filter may be used. In this embodiment, an example using the overlap storage method is shown, but an overlap addition method may be used. In the present embodiment, an example using FFT conversion is shown, but MDCT conversion may be used.

The signal selection unit (SEL) 428 includes residual signals e ₁ [n] = (e ₁ [nF−F + 1], e ₁ [nF−F + 2],... Output from the first echo canceller unit (EC1) 4271. .., E ₁ [nF]) and the residual signal e ₂ [n] = (e ₂ [nF−F + 1], e ₂ [nF−F + 2] output from the second echo canceller (EC2) 4272. ,..., E ₂ [nF]) as input and the transmission output signal s ′ [n] = (s ′ [nF−F + 1], s ′ [nF−F + 2],. ]) Is calculated and output to the communication unit (COM) 11.

Specifically, the signal selecting unit and the (SEL) 428, the power of the residual signal _e 1 of frame [n] (2 sum of squares of F samples of the residual signal _e 1 [n].), Frame Is compared with the power of the residual signal e ₂ [n] (the sum of squares of F samples of the residual signal e ₂ [n]), and the signal with the lower power is sent as is to the transmission output signal s ′ [ n].

  Here, since the adaptive algorithm cannot perform echo suppression exceeding the background noise level, the two residual signals do not become unnatural sounds in which background noise varies due to echo suppression processing. Further, by selecting the two residual signals by power comparison in this way, a signal with more echo suppression can be output as a transmission output signal.

  A processing flow of the signal processing device according to the fourth embodiment configured as described above will be described with reference to FIGS. 22 to 23. FIG. 22 is a flowchart showing a processing flow of the entire signal processing apparatus, and FIG. 23 is a flowchart showing a processing flow in the first echo canceller unit (EC1) 4271 and the second echo canceller unit (EC2) 4272. . In addition, about the same operation step as the operation | movement of the signal processing apparatus which concerns on 1st Embodiment demonstrated with reference to FIGS. 7-9, the same code | symbol is attached | subjected and description of the part is abbreviate | omitted.

In FIG. 22, after step S1002, as a buffer process, the reception input signal x _B0 [n] output from the reception signal storage unit (RBUFF) 421 is input, and the reception signal delay processing unit (RDP) 4251 receives the reception input signal x. _B1 [n] is output, and the reception signal delay processing unit (RDP) 4252 outputs the reception input signal x _B2 [n] (step S4004).

The first echo canceller (EC1) 4271 receives the received input signal x _B1 [n] and the transmitted input signal z [n], and the second echo canceller (EC2) 4272 receives the received input signal x _B2. Echo canceller processing is performed with [n] and the transmission input signal z [n] as inputs (step S4005).

The signal selection unit (SEL) 428 includes a residual signal e ₁ [n] output from the first echo canceller unit (EC1) 4271 and a residual signal output from the second echo canceller unit (EC2) 4272. Using e ₂ [n] as an input, the transmission output signal s ′ [n] is calculated and output (step S4007).

  Then, according to the determination processing in step S1006, the processing from step S1002 to step S4007 is performed until the call ends.

  The processing of the first echo canceller (EC1) 4271 and the second echo canceller (EC2) 4272 shown in FIG. 23 is performed as follows.

First, the frequency domain transform processing unit (FT) 4271a converts the received input signal x _B1 [n] output from the received signal delay processing unit (RDP) 4251 into the frequency domain by FFT transform, and the frequency of the received input signal. A spectrum X ₁ [n, ω] is calculated.

Next, the frequency domain adaptive filter unit (FDAF) 4271b uses the frequency spectrum X ₁ [n, ω] of the received input signal and the filter coefficient H ₁ [n, ω] to generate the frequency spectrum Y ′ ₁ [ n, ω] is calculated. Then, the frequency domain inverse transform processing unit (IFT) 4271c performs IFFT conversion on the frequency spectrum Y ′ ₁ [n, ω] of the pseudo echo signal to calculate the pseudo echo signal y ′ ₁ [n]. Then, the signal subtraction processing unit 4271d subtracts the pseudo echo signal y ′ ₁ [n] from the transmission input signal z [n] to calculate a residual signal e ₁ [n] (step S4201-1).

Similarly, the frequency domain conversion processing unit (FT) 4272a converts the reception input signal x _B2 [n] output from the reception signal delay processing unit (RDP) 4252 into the frequency domain by FFT conversion, and receives the reception input signal. Frequency spectrum X ₂ [n, ω] is calculated.

Next, the frequency domain adaptive filter unit (FDAF) 4272b uses the frequency spectrum X ₂ [n, ω] of the received input signal and the filter coefficient H ₂ [n, ω] to generate the frequency spectrum Y ′ ₂ [ n, ω] is calculated. Then, the frequency domain inverse transform processing unit (IFT) 4272c performs IFFT conversion on the frequency spectrum Y ′ ₂ [n, ω] of the pseudo echo signal to calculate the pseudo echo signal y ′ ₂ [n]. Then, the signal subtraction processing unit 4272d subtracts the pseudo echo signal y ′ ₂ [n] from the transmission input signal z [n] to calculate a residual signal e ₂ [n] (step S4201-2).

Next, the frequency domain transform processing unit (FT) 4271e transforms the residual signal e ₁ [n] into the frequency domain by FFT transform, and the frequency domain adaptive filter control unit (FDCNT) 4271f stops learning as an adaptive filter control process. Control information EC1state [n, ω], which is information indicating whether or not the state is present, is output (step S4202-1).

Similarly, the frequency domain transform processing unit (FT) 4272e transforms the residual signal e ₂ [n] into the frequency domain by FFT transform, and the frequency domain adaptive filter control unit (FDCNT) 4272f stops learning as an adaptive filter control process. Control information EC2state [n, ω], which is information indicating whether or not the state is present, is output (step S4202-2).

The frequency domain adaptive filter unit (FDAF) 4271b receives the control of the control information EC1state [n, ω] output from the frequency domain adaptive filter control unit (FDCNT) 4271f, and the filter coefficient H ₁ [n, ω]. Is subjected to adaptive learning processing (step S4203-1).

Similarly, the frequency domain adaptive filter unit (FDAF) 4272b receives the control of the control information EC2state [n, ω] output from the frequency domain adaptive filter control unit (FDCNT) 4272f, and the filter coefficient H ₂ [n, ω ] Is adaptively learned (step S4203-2), and the echo canceller process ends.

  In the above, there are two echo canceller units, the first echo canceller unit (EC1) 4271 and the second echo canceller unit (EC2) 4272, but when the number q of echo canceller units is q> 2, It is configured and operates in the same manner as described above.

  In the above description, the first echo canceller unit (EC1) 4271 and the second echo canceller unit (EC2) 4272 are frequency domain type adaptive systems without gradient constraints as a method for performing high-speed processing of block processing adaptive filters using FFT transform. Although an example of a filter is shown, it may be configured by a frequency domain adaptive filter with a gradient constraint.

  Further, the first echo canceller (EC1) 4271 and the second echo canceller (EC2) 4272 show an example of the overlap preserving method using FFT transform in this embodiment, but the overlap addition method Or MDCT conversion may be used.

Further, the first echo canceler unit (EC1) 4271 and the second echo canceller unit (EC2) 4272 have the filter length N and the actual echo length (an echo effective for echo suppression), as in this embodiment. When the length is set to L, the received signal corresponding to the past by the time shift width M _D samples satisfying 1 ≦ M _D ≦ N−L from the received signal input from the received signal delay processor (RDP) 4251 By outputting the input signal from the received signal delay processing unit (RDP) 4252, a time domain type sample processing adaptive filter may be used.

By the operation of the signal processing apparatus according to the fourth embodiment described above, reception input signals having different delay amounts are input to the first echo canceller unit (EC1) 4271 and the second echo canceller unit (EC2) 4272. Then, the echo suppression processing is performed, and the result of the echo suppression processing is selected according to a predetermined selection criterion by the signal selection unit (SEL) 428, so that the delay amount is not estimated. the can not is dependent, stable echo suppression performance can be obtained even when the true delay D ^* variation can be made robust to true delay amount D ^* variation.

  In the above description, the echo suppression processing unit (ECP) 42 includes two echo canceller units, that is, a first echo canceller unit (EC1) 4271 and a second echo canceller unit (EC2) 4272. Although mainly, it may have three or more echo cancellers.

  In the case of having three or more echo canceller units, the input of each echo canceller unit is a received input signal that goes back in a different time in the past. For example, an echo suppression processing unit (ECP) 42 is connected to each echo canceller unit. It is output from the corresponding received signal delay processing unit. Further, the signal selection unit (SEL) 428 compares the powers of the residual signals output from the respective echo canceller units, and sets the residual signal having the lowest power as the transmission output signal s ′ [n].

By having a plurality of echo canceller units, and receiving the input of each echo canceller unit as a received input signal that dates back in the past, a received input signal corresponding to the true delay amount D ^* in any echo canceller unit it is possible to enter a stable echo suppression performance can be obtained even when the true delay D ^* variation can be made robust to true delay amount D ^* variation.

(Modification of the fourth embodiment)
The signal processing apparatus according to the modification of the fourth embodiment of the present invention differs from the signal processing apparatus according to the fourth embodiment in that an echo suppression processing unit (ECP) is used instead of the echo suppression processing unit (ECP) 42. ) 422. FIG. 24 is a block diagram illustrating a configuration of an echo suppression processing unit (ECP) 422 of a signal processing device according to a modification of the fourth embodiment of the present invention.

  The difference between the echo suppression processing unit (ECP) 422 and the echo suppression processing unit (ECP) 42 is that the frequency domain conversion processing unit (FT) 4271a and the second echo canceller unit (EC1) 4271 of the first echo canceller unit (EC1) 4271 ( EC2) The frequency domain transform processing unit (FT) 4272a is replaced with a frequency domain transform processing unit (FT) 427a, and the received signal delay processing unit (RDP) 4252 is not included.

  Further, the output of the frequency domain transform processing unit (FT) 427a is converted from the frequency domain adaptive filter unit (FDAF) 4271b and the frequency domain adaptive filter control unit (FDCNT) 4271f of the first echo canceller unit (EC1) 4271, and the second The echo canceller unit (EC2) 4272 is input to the frequency domain adaptive filter unit (FDAF) 4272b and the frequency domain adaptive filter control unit (FDCNT) 4272f, and the other parts are the same.

  Therefore, in the echo suppression processing unit (ECP) 422, the same parts as those of the echo suppression processing unit (ECP) 42 are denoted by the same reference numerals and description thereof is omitted.

The frequency domain transform processing unit (FT) 427a receives the reception input signal x _B1 [n] = (x [nF-2B + 1], x [nF-2B + 2] for 2B samples output from the reception signal delay processing unit (RDP) 4251. ],..., X [nF]) as input, and the frequency spectrum X ₁ [n, ω] of the received input signal is calculated by 2B-point FFT conversion to calculate the first echo canceller. Output to the frequency domain adaptive filter unit (FDAF) 4271b and the frequency domain adaptive filter control unit (FDCNT) 4271f of the unit (EC1) 4271.

In addition, the frequency domain transform processing unit (FT) 427a sets the frequency spectrum X ₁ [n−1, ω] of the received input signal one frame before as X ₂ [n, ω], and the second echo canceler unit (EC2). ) 4272 to the frequency domain adaptive filter unit (FDAF) 4272b and the frequency domain adaptive filter control unit (FDCNT) 4272f.

Here, the frequency domain transform processing unit (FT) 427a outputs the frequency spectrum X ₁ [n, ω] of the received input signal and the frequency spectrum X ₁ [n−1, ω] of the received input signal one frame before. Thus, it is possible to output a reception input signal having a different delay amount by one frame to the first echo canceller (EC1) 4271 and the second echo canceller (EC2) 4272.

In this embodiment, the frequency domain transform processing unit (FT) 427a outputs the frequency spectrum X ₁ [n, ω] of the received input signal and the frequency spectrum X ₁ [n−1, ω] of the received input signal one frame before. An example to do. When actual echo length (effective echo length echo suppression) is L, the time shift width M _D of the two received input signals need only shift M _D samples satisfying _{1 ≦ M D ≦ B-L} , Here, an example corresponding to a delay amount for one frame, that is, M _D = F is shown.

Since the time shift width M _D of the two received input signals may be a multiple of F satisfies the _{1 ≦ M D ≦ B-L} , the frequency domain transform processor when (FT) 427a is several frames before received input The frequency spectrum of the signal may be output.

  Due to the operation of the signal processing apparatus according to the modification of the fourth embodiment described above, the echo suppression processing unit (ECP) 422 is replaced with the frequency in place of the echo suppression processing unit (ECP) 42 compared to the fourth embodiment. Since the area conversion processing unit (FT) 4272a is not provided, the calculation amount can be reduced.

  The echo suppression processing unit (ECP) 422 includes two echo canceller units, that is, a first echo canceller unit (EC1) 4271 and a second echo canceller unit (EC2) 4272. The echo canceller may be provided.

  When there are three or more echo canceller units, the input of each echo canceller unit is the frequency spectrum of the received input signal that goes back by one frame in the past, for example, output from the frequency domain transform processing unit (FT) 427a. Is done. The operation of the signal selection unit (SEL) 428 is as described above.

By having a plurality of echo canceller units, and receiving the input of each echo canceller unit as a received input signal that dates back in the past, a received input signal corresponding to the true delay amount D ^* in any echo canceller unit it is possible to enter a stable echo suppression performance can be obtained even when the true delay D ^* variation can be made robust to true delay amount D ^* variation.

(Fifth embodiment)
The signal processing device according to the fifth embodiment is different from the signal processing device according to the first embodiment in that it does not have the echo suppression processing unit (ECP) 12 but has an echo suppression processing unit (ECP) 52. The other parts are the same. Therefore, the same parts are denoted by the same reference numerals and the description thereof is omitted.

  FIG. 25 is a block diagram showing a configuration of an echo suppression processing unit (ECP) 52 of the signal processing device according to the fifth embodiment of the present invention. The echo suppression processing unit (ECP) 52 includes a reception signal storage unit (RBUFF) 521 connected to the communication unit (COM) 11, a reception signal delay processing unit (RDP) 5251, and a reception signal delay processing unit (RDP). 5252, a first echo reduction unit (ER1) 5271, a second echo reduction unit (ER2) 5272, and a frequency domain conversion processing unit (FT) connected to the A / D converter (A / D) 18 527b, transmission power calculation unit (POW) 527d, frequency domain inverse transform processing unit (IFT) 527L connected to communication unit (COM) 11, signal selection unit (SEL) 528, noise estimation unit (NE) ) 529.

  FIG. 26 is a block diagram showing a configuration of a Qth echo reduction unit (ERQ) 527Q (where Q is 1 or 2). The Q-th echo reduction unit (ERQ) 527Q includes a frequency domain transform processing unit (FT) 527Qa connected to the reception signal delay processing unit (RDP) 525Q, a reception power calculation unit (POW) 527Qc, and transmission power calculation. An acoustic coupling amount estimation unit (ACLE) 527Qe connected to the unit (POW) 527d, an echo amount estimation unit (ELE) 527Qf, a frequency domain control unit (FCNT) 527Qg, a gain storage unit (GTBL) 527Qh, An echo suppression gain calculation unit (GCAL) 527Qi connected to the speech power calculation unit (POW) 527d, and a signal suppression unit (SS) 527Qj connected to the signal selection unit (SEL) 528.

The window function w [i used in the frequency domain transform processing unit (FT) 527b and the frequency domain transform processing unit (FT) 527Qa (where Q is 1 or 2) of the Qth echo reduction unit (ERQ) 527Q. ] (I = 0, 1,..., L _F ) are obtained by the frequency domain transform processing unit (FT) 327a and the frequency domain transform processing unit (FT) 327b according to the third embodiment shown in FIG. Similar to the window function used. When the effective window length of the window function w [i] is L _W (length except when the window function w [i] is 0), and the overlap of the window function w [i] is L _O , 1 ≦ _{M D} ≦ _L W satisfy -L _O setting time shift width _{M D} in advance.

In this embodiment, there are two echo reduction units, a first echo reduction unit (ER1) 5271 and a second echo reduction unit (ER2) 5272. The number of echo reduction units q and the received input signal x [n ] Is set so that the relationship of D _MAX ≦ q · M _D is established in the maximum fluctuation width D _MAX of the true delay amount D ^* between the echo and the transmission input signal z [n]. The number q of echo reduction units is set according to the value of _D.

  The operation of each part of the echo suppression processing unit (ECP) 52 of the signal processing device according to the fifth embodiment of the present invention configured as described above will be described with reference to FIG. 25 and FIG.

The reception signal storage unit (RBUFF) 521 receives the reception input signal x [n] for each frame output from the communication unit (COM) 11, and receives the first echo reduction unit (ER1) 5271 and the second echo. refer to a valid window length _{L W} of the reduction unit (ER2) 5272 window function _{w [i], L W +} M D samples of the received input signal _{x W0 [n] = (x} [nF-L W -M _D + 1], x [nF−L _W −M _D +2],..., X [nF]) are stored and output.

Received signal delay processing unit (RDP) 5251 is the received signal storage unit (RBUFF) _L outputted from the 521 W + _{M D} samples of the received input signal _x W0 [n] as input, _{L W} samples of the received input signal x _W1 [n] = (x [nF−L _W +1], x [nF−L _W +2],..., x [nF]) is output.

From the received signal delay processor (RDP) 5252 is the received signal storage unit (RBUFF) _L outputted from the 521 W + _{M D} samples of the received input signal _x W0 [n] as _input, among _x W0 [n] Receive input signal x _W2 [n] = (x [nF−L _W −M _D +1], x [nF−L _W −M _D +2],..., X for L _W samples corresponding to the past time [NF-M _D ]) is output. That, and outputs the received input signal corresponding to the _{M D} samples past than received signal delay processor (RDP) received input signal output from the 5251 _x W1 [n] _x W2 as [n].

The frequency domain transform processing unit (FT) 527b receives the transmission input signal z [n] output from the A / D converter (A / D) 18, and uses the frequency spectrum Z [n, ω of the transmission input signal. ] Is calculated and output. Specifically, using L _W samples of the transmission input signal z [n], also used in conjunction one frame past of the transmission input signal z [n-1], and the like zero-padded L _F samples generates a signal, this window function w [i] in the _{(i = 0,1, ···, L} F) to windowing and converted to the frequency domain by the FFT transformation of _{L F} point, the sending input The frequency spectrum Z [n, ω] of the signal is calculated and output.

The transmission power calculation unit (POW) 527d receives the frequency spectrum Z [n, ω] of the transmission input signal output from the frequency domain conversion processing unit (FT) 527b as an input, and the transmission power spectrum that is the power spectrum thereof. | Z [n, ω] | ² is calculated. Since the amount of acoustic coupling usually does not change suddenly in time, the amount of acoustic coupling can be estimated more accurately by using a smoothed value than by using an instantaneous value. Therefore, the transmission power calculation unit (POW) 527d For example, as shown in Equation 20, the transmission power spectrum | Z _S [n, ω] | ² smoothed using the value | Z _S [n−1, ω] | ² of the previous frame is calculated and output. To do. However, α _Z [ω] is preferably about 0.75 to 0.999.

The first echo reduction unit (ER1) 5271 is output from the reception input signal x _W1 [n] output from the reception signal delay processing unit (RDP) 5251 and the A / D converter (A / D) 18. The transmission input signal z [n] is used as an input, and the spectrum of the echo-suppressed signal obtained by suppressing the echo component from the transmission input signal z [n] is output every frame as S ′ ₁ [n, ω]. .

The second echo reduction unit (ER2) 5272 outputs the received input signal x _W2 [n] output from the received signal delay processing unit (RDP) 5252 and the A / D converter (A / D) 18. The spectrum of the echo-suppressed signal obtained by suppressing the echo component from the transmission input signal z [n] is S ′ ₂ [n, ω] for each frame. Output.

  Next, with reference to FIG. 26, the operation of each unit constituting the Qth echo reduction unit (ERQ) 527Q (where Q is 1 or 2) will be described.

The frequency domain transform processing unit (FT) 527Qa receives the reception input signal x _WQ [n] output from the reception signal delay processing unit (RDP) 525Q and inputs the frequency spectrum X _Q [n, ω] of the reception input signal. Calculate and output. Specifically, the reception input signal x _WQ [n] for L _W samples is used, and the reception input signal x _WQ [n−1] of one frame in the past is used together, and a signal for _LF samples is obtained by zero padding or the like. generated, this window function w [i] (i = 0,1 , ···, L F) to windowing and converted to the frequency domain by the FFT transformation of _{L F} point, the frequency of the received input signal The spectrum X _Q [n, ω] is calculated and output.

The received power calculation unit (POW) 527Qc receives the frequency spectrum X _Q [n, ω] of the received input signal output from the frequency domain transform processing unit (FT) 527Qa as an input, and the received power spectrum | X _Q [n, ω] | ² is calculated. Since the acoustic coupling amount does not usually change suddenly in time, the acoustic coupling amount can be estimated with higher accuracy by using a smoothed value than by using an instantaneous value. Therefore, the received power calculation unit (POW) 527Qc is For example, as shown in Equation 21, the received power spectrum | X _QS [n, ω] | ² smoothed using the value | X _QS [n−1, ω] | ² of the previous frame is calculated and output. However, α _QX [ω] is preferably about 0.75 to 0.999.

The acoustic coupling amount estimation unit (ACLE) 527Qe includes a smoothed reception power spectrum | X _QS [n, ω] | ² output from the reception power calculation unit (POW) 527Qc, and a transmission power calculation unit (POW) 527d. , The smoothed transmission power spectrum | Z _S [n, ω] | ² output from the control information ERQstate [n, ω] output from the frequency domain control unit (FCNT) 527Qg, and the frequency band For each ω, the acoustic coupling amount | H _Q [n, ω] | ² is calculated by, for example, the following Expression 22.

Then, the acoustic coupling amount estimation unit (ACLE) 527Qe calculates and outputs a smoothed acoustic coupling amount | H _QS [n, ω] | ² using a value one frame before as in the following Expression 23. However, α _QH [ω] is preferably about 0.03 to 0.99.

Here, at the beginning of the call, for example, by increasing α _QH [ω] for about five seconds from the start of the call, the update of the acoustic coupling amount | H _QS [n, ω] | ² is accelerated. By doing so, since the acoustic coupling amount is initialized at the beginning of the call, it is possible to prevent the suppression amount from being reduced at the beginning of the call.

However, when the control information ERQstate [n, ω] output from the frequency domain control unit (FCNT) 527Qg indicates that the learning is stopped, the acoustic coupling amount estimation unit (ACLE) 527Qe does not update the acoustic coupling amount. Then, the past acoustic coupling amount | H _QS [n−1, ω] | ² is used.

The echo amount estimation unit (ELE) 527Qf includes the smoothed reception power spectrum | X _QS [n, ω] | ² output from the reception power calculation unit (POW) 527Qc and the acoustic coupling amount estimation unit (ACLE) 527Qe. The output acoustic coupling amount | H _QS [n, ω] | ² is input, and the echo amount | Y _Q [n, ω] | ² contained in the frequency spectrum Z [n, ω] of the transmission input signal is Estimation is performed for each frequency band ω as shown in Equation 24 below.

Then, the echo amount estimation unit (ELE) 527Qf can make the signal after echo suppression more natural by using the smoothed value than by using the instantaneous echo amount | Y _Q [n, ω] | ^2. Therefore, the echo amount | Y _QS [n, ω] | ² smoothed using the value of the previous frame as in the following Expression 25 is calculated and output for each frequency band ω. However, α _QY [ω] is preferably about 0.7 to 0.99.

The frequency domain control unit (FCNT) 527Qg includes the smoothed reception power spectrum | X _QS [n, ω] | ² output from the reception power calculation unit (POW) 527Qc and the acoustic coupling amount estimation unit (ACLE) 527Qe. acoustic coupling of one frame before output _{| H QS [n-1,} ω] | ² and an input, the control information is information indicating whether the learning stop state for each frame n and frequency bands omega ERQstate [n, ω] is output.

Specifically, the frequency domain control unit (FCNT) 527Qg has a case where the amount of acoustic coupling changes rapidly, that is, | H _QS [n, ω] | ² > β _QH [ω] · | H _QS [n− 1, ω] | ² or when the received input signal is not sufficiently large, that is, when | X _QS [n, ω] | ² <β _QX [ω] is satisfied. It is determined that On the other hand, if none of these cases apply, it is determined that the learning is not stopped. However, β _QH [ω] is preferably about 0.9 to 30. β _QX [ω] is preferably about 30 dB to 40 dB.

The gain storage unit (GTBL) 527Qh stores and outputs a parameter γ _Q [ω] that controls a preset nonlinear echo suppression amount. However, γ _Q [ω] is preferably about 1.0 to 2.0.

The echo suppression gain calculation unit (GCAL) 527Qi includes a smoothed transmission power spectrum | Z _S [n, ω] | ² output from the transmission power calculation unit (POW) 527d, and an echo amount estimation unit (ELE). The smoothed echo amount | Y _QS [n, ω] | ² output from 527Qf and the parameter γ _Q [ω] output from the gain storage unit (GTBL) 527Qh are input, and the echo suppression gain G _Q [n , Ω] is calculated and output.

Specifically, the echo suppression gain calculation unit (GCAL) 527Qi calculates the echo suppression gain G _Q [n, ω] according to Equation 26 using the Wiener filter method.

The signal suppression unit (SS) 527Qj outputs the frequency spectrum Z [n, ω] of the transmission input signal output from the frequency domain transform processing unit (FT) 527b and the echo suppression gain calculation unit (GCAL) 527Qi. Using the echo suppression gain G _Q [n, ω] as an input, the echo of the frequency spectrum Z [n, ω] of the transmission input signal is suppressed, and the spectrum S ′ of the transmission output signal is expressed as shown in Equation 27 below. Output as _Q [n, ω].

  Referring back to FIG. 25, the description returns to the operation of each unit constituting the echo suppression processing unit (ECP) 52.

The noise estimation unit (NE) 529 receives the frequency spectrum Z [n, ω] of the transmission input signal output from the frequency domain conversion processing unit (FT) 527b as input, and the sound of the transmission input signal z [n] The spectrum of a section that is not a section is measured, the noise level | N [n, ω] | ² included in the frequency spectrum is estimated for each frequency band ω, and the noise frequency spectrum N [n, ω] is calculated and output. To do.

The signal selection unit (SEL) 528 includes the spectrum S ′ ₁ [n, ω] of the transmission output signal output from the signal suppression unit (SS) 5271j of the first echo reduction unit (ER1) 5271, and the second The spectrum S ′ ₂ [n, ω] of the transmission output signal output from the signal suppression unit (SS) 5272j of the echo reduction unit (ER2) 5272 and the frequency spectrum of the noise output from the noise estimation unit (NE) 529 Using N [n, ω] as an input, the frequency spectrum S ′ [n, ω] of the transmission output signal is calculated and output.

Specifically, the power spectrums | S ′ ₁ [n, ω] | ² and | S ′ ₂ [n, ω] | ² of the transmission output signal are larger than the noise level | N [n, ω] | ^2. when the signal selector (SEL) 528, the power spectrum of the transmitter output signal _{| S '1 [n, ω} ] | 2 and _{| S' 2 [n, ω} ] | frequency spectrum of the ² who power is small Is selected for each frequency band ω.

If the power spectrums | S ′ ₁ [n, ω] | ² and | S ′ ₂ [n, ω] | ^{2 of} the transmission output signal are smaller than the noise level | N [n, ω] | ² , The signal selection unit (SEL) 528 selects the frequency spectrum N [n, ω] of noise so that background noise is not excessively suppressed by echo suppression processing and an unnatural sound is not generated.

Alternatively, the signal selection unit (SEL) 528, when the transmission input signal z [n] is a voiced section, the power spectrum | S ′ ₁ [n, ω] | ² and | S ′ of the transmission output signal. ₂ [n, ω] | ² , the frequency spectrum with the smaller power is selected for each frequency band ω, and when the transmission input signal z [n] is not a voiced section, the noise frequency spectrum N [n, [omega]] may be selected.

  The frequency domain inverse transform processing unit (IFT) 527L receives the frequency spectrum S ′ [n, ω] of the transmission output signal output from the signal selection unit (SEL) 528 as an input, and the transmission output signal s ′ [n]. = (S ′ [nF−F + 1], s ′ [nF−F + 2],..., S ′ [nF]) is calculated and output to the communication unit (COM) 11.

Specifically, the frequency domain inverse transform unit (IFT) 527L, the frequency spectrum S '[n, ω] of the transmitter output signal is converted into a time region by IFFT conversion of _{L F} point, frequency domain transform processor Referring to the windowing by the window function w [i] (i = 0, 1,..., L _F ) of the part (FT) 527b, the overlap L _O samples are appropriately transmitted output signals of past frames. The process of returning the overlap is performed by superimposing s ′ [n] on the transmission output signal s ′ [n] of the current frame, and the transmission output signal s ′ [n] = (s ′ [nF−F + 1]. , S ′ [nF−F + 2],..., S ′ [nF]) are calculated and output.

  A processing flow of the signal processing device according to the fifth embodiment configured as described above will be described with reference to FIGS. FIG. 27 is a flowchart showing a processing flow of the entire signal processing apparatus, and FIG. 28 is a flowchart showing a processing flow in the first echo reduction unit (ER1) 5271 and the second echo reduction unit (ER2) 5272. . The same operations as those of the signal processing device according to the first embodiment described with reference to FIGS. 7 to 9 and the signal processing device according to the third embodiment described with reference to FIGS. About a step, the same code | symbol is attached | subjected and description of the part is abbreviate | omitted.

After step S1002 in FIG. 27, as buffering, and input the reception signal storage unit (RBUFF) 521 received input signal _x w0 output from [n], the received signal delay processing unit (RDP) 5251 is received input signal _{x w1} [N] is output, and the reception signal delay processing unit (RDP) 5252 outputs the reception input signal x _w2 [n] (step S5004).

The first echo reduction unit (ER1) 5271 receives the received input signal x _w1 [n] and the transmitted input signal z [n], and the second echo reduction unit (ER2) 5272 receives the received input signal x _w2. Echo reduction processing is performed using [n] and the transmission input signal z [n] as inputs (step S5005).

The signal selection unit (SEL) 528 includes the spectrum S ′ ₁ [n, ω] of the transmission output signal output from the first echo reduction unit (ER1) 5271 and the second echo reduction unit (ER2) 5272. Using the spectrum S ′ ₂ [n, ω] of the output transmission output signal as an input, the frequency spectrum S ′ [n, ω] of the transmission output signal is calculated and output (step S5007).

  The frequency domain inverse transform processing unit (IFT) 527L receives the frequency spectrum S ′ [n, ω] of the transmission output signal output from the signal selection unit (SEL) 528 as an input, and transmits the transmission output signal s ′ [ n] is calculated and output (step S5008).

  Then, according to the determination processing in step S1006, the processing from step S1002 to step S5008 is performed until the call ends.

The processes of the first echo reduction unit (ER1) 5271 and the second echo reduction unit (ER2) 5272 shown in FIG. 28 are performed as follows. First, the frequency domain transform processing unit (FT) 5271a, the frequency domain transform processing unit (FT) 5272a, and the frequency domain transform processing unit (FT) 527b respectively receive the received input signal x _W1 [n] for L _W samples, A signal x _W2 [n] and a transmission input signal z [n] are used to generate a signal for _LF samples by using a signal of one frame past or zero-padded, and converting it to the frequency domain The reception frame and transmission frame, which are buffers, are respectively updated (steps S5201r-1, S5201r-2, and S5201s).

Next, frequency domain transform processor (FT) 5271a, and windowing a window function w [i] in _{L F} samples of the signal frequency domain transform processor (FT) 5272a is stored in the reception frame, L The frequency spectrum is converted to the frequency domain by _F- point FFT conversion, and the frequency spectra X ₁ [n, ω] and X ₂ [n, ω] of the received input signal are calculated (steps S5202r-1 and S5202r-2). The received power calculation unit (POW) 5271c, received power calculation unit (POW) 5272c respectively received power spectrum ^{_{| X 1 [n, ω]}} | 2, | X 2 [n, ω] | 2 and the smoothed received The power spectra | X _1S [n, ω] | ² and | X _2S [n, ω] | ² are calculated (steps S5203r-1 and S5203r-2).

On the other hand, the calculation of the frequency spectrum Z [n, ω] by the frequency domain transform processing unit (FT) 527b in step S3202s and the transmission power spectrum | Z _S [n, ω by the transmission power calculation unit (POW) 527d in step S3203s. ] | ² is calculated.

  Then, following the processing in step S5203r-1, the processing in step S5203r-2, and the processing in step S3203s, the frequency domain control unit (FCNT) 5271g outputs control information ER1state [n, ω], and the frequency domain control unit ( FCNT) 5272g outputs control information ER2state [n, ω].

Subsequently, the acoustic coupling amount estimating unit (ACLE) 5271e is smoothed received power spectrum _{| X 1S [n, ω]} | 2 and smoothed sending power spectrum _{| Z S [n, ω]} | ^2, An acoustic coupling amount | H _1S [n, ω] | ² is calculated using the control information ER1state [n, ω] as an input (step S5204-1).

Similarly, the acoustic coupling amount estimating unit (ACLE) 5272e is smoothed received power spectrum _{| X 2S [n, ω]} | 2 and smoothed sending power spectrum _{| Z S [n, ω]} | ^2, The acoustic coupling amount | H _2S [n, ω] | ² is calculated using the control information ER2state [n, ω] as an input (step S5204-2).

Thereafter, the echo estimation unit (ELE) 5271f an acoustic coupling amount _{| H 1S [n, ω]} | 2 and smoothing the received power spectrum _{| X 1S [n, ω]} | 2 and the transmission input signal as an input The amount of echo contained | Y _1S [n, ω] | ² is estimated (step S5205-1). Similarly, the echo estimation unit (ELE) 5272f an acoustic coupling amount _{| H 2S [n, ω]} | 2 and smoothing the received power spectrum _{| X 2S [n, ω]} | 2 and transmission input signal as an input The amount of echo | Y _2S [n, ω] | ^{2 included in} is estimated (step S5205-2).

Next, the echo suppression gain calculation unit (GCAL) 5271i includes the smoothed transmission power spectrum | Z _S [n, ω] | ² output from the transmission power calculation unit (POW) 527d, and the echo amount estimation unit. (ELE) Using the smoothed echo amount | Y _1S [n, ω] | ² output from the 5271f and the parameter γ ₁ [ω] output from the gain storage unit (GTBL) 5271h as inputs, the echo suppression gain G ₁ [n, ω] is calculated (step S5206-1).

Similarly, the echo suppression gain calculation unit (GCAL) 5272i includes the smoothed transmission power spectrum | Z _S [n, ω] | ² output from the transmission power calculation unit (POW) 527d, and the echo amount estimation unit. (ELE) The smoothed echo amount | Y _2S [n, ω] | ² output from the 5272f and the parameter γ ₂ [ω] output from the gain storage unit (GTBL) 5272h are input, and the echo suppression gain G ₂ [n, ω] is calculated (step S5206-2).

Finally, the signal suppression unit (SS) 5271j transmits the frequency spectrum Z [n, ω] of the transmission input signal and the echo suppression gain G ₁ [n, ω] calculated by the echo suppression gain calculation unit (GCAL) 5271i. Using [ω] as an input, the frequency spectrum S ′ ₁ [n, ω] of the transmission output signal in which the echo component of the frequency spectrum Z [n, ω] of the transmission input signal is suppressed is output (step S5207-1).

Similarly, the signal suppression unit (SS) 5272j includes the frequency spectrum Z [n, ω] of the transmission input signal and the echo suppression gain G ₂ [n, ω] calculated by the echo suppression gain calculation unit (GCAL) 5272i. Is input, and the frequency spectrum S ′ ₂ [n, ω] of the transmission output signal in which the echo component of the frequency spectrum Z [n, ω] of the transmission input signal is suppressed is output (step S5207-2), and the echo is output. The reduction process ends.

  In the above, there are two echo reduction units, the first echo reduction unit (ER1) 5271 and the second echo reduction unit (ER2) 5272, but when the number q of echo reduction units is q> 2, It is configured and operates in the same manner as described above.

  In the above description, the first echo reduction unit (ER1) 5271 and the second echo reduction unit (ER2) 5272 have been described as operating as a method of processing for each frequency band by frequency domain conversion by FFT conversion. You may implement | achieve the frequency domain type | mold echo suppression processing, such as the method which puts together the frequency band by FFT conversion, and processes for every frequency band group, the method using MDCT conversion, and band division filters, such as a filter bank.

  By the operation of the signal processing apparatus according to the fifth embodiment described above, received input signals having different delay amounts are input to the first echo reduction unit (ER1) 5271 and the second echo reduction unit (ER2) 5272. Echo suppression processing. Then, as long as the power spectrum is larger than the noise level estimated by the noise estimation unit (NE) 529, the signal selection unit (SEL) 528 uses the minimum power from the frequency spectrum of the signals after the echo suppression processing as a reference. Select one signal.

According to this processing, since the delay amount is not estimated, the echo suppression processing performance can be made independent of the delay amount estimation accuracy, and stable echo suppression processing can be performed even when the true delay amount D ^* varies. Performance can be obtained, robust against fluctuations in the true delay amount D ^* , preventing intermittent suppression of background noise, and preventing deterioration of transmitted voice quality due to excessive echo suppression can do.

  In the above description, the echo suppression processing unit (ECP) 52 includes two echo reduction units, that is, a first echo reduction unit (ER1) 5271 and a second echo reduction unit (ER2) 5272. Although mainly, it may have three or more echo reduction units.

  In the case of having three or more echo reduction units, the input of each echo reduction unit is a received input signal that goes back in the past in different times, and for example, an echo suppression processing unit (ECP) 52 is connected to each echo reduction unit. It is output from the corresponding received signal delay processing unit.

  In the case of having three or more echo reduction units, the signal selection unit (SEL) 528 outputs the power spectrum of the transmission output signal output from each echo reduction unit from the noise estimation unit (NE) 529. If it is greater than the noise level, the signal having the smallest power in the power spectrum of the transmission output signal is selected for each frequency band ω.

  On the other hand, when the power spectrum of the transmission output signal is smaller than the noise level output from the noise estimation unit (NE) 529, the signal selection unit (SEL) 528 selects the noise frequency spectrum N [n, ω]. .

  Alternatively, the signal selection unit (SEL) 528 selects a signal having the lowest power in the power spectrum of the transmission output signal for each frequency band ω when the transmission input signal z [n] is a sound section. When the transmission input signal z [n] is not a sound section, the noise frequency spectrum N [n, ω] may be selected.

By having a plurality of echo canceller units, and receiving the input of each echo canceller unit as a received input signal that dates back in the past, a received input signal corresponding to the true delay amount D ^* in any echo canceller unit it is possible to enter a stable echo suppression performance can be obtained even when the true delay D ^* variation can be made robust to true delay amount D ^* variation.

(Sixth embodiment)
The signal processing device according to the sixth embodiment is different from the signal processing device according to the modification of the fourth embodiment in that the echo suppression processing unit (ECP) 422 is not provided and the echo suppression processing unit (ECP) 62 is not provided. The other parts are the same. Therefore, the same parts are denoted by the same reference numerals and the description thereof is omitted. Also, it is a part constituting the echo suppression processing unit (ECP) 62 and the same part as the part constituting the echo suppression processing unit (ECP) 422 according to the modification of the fourth embodiment is denoted by the same reference numeral. A description thereof will be omitted.

  FIG. 29 is a block diagram showing a configuration of an echo suppression processing unit (ECP) 62 of the signal processing apparatus according to the sixth embodiment of the present invention. The echo suppression processing unit (ECP) 62 includes a reception signal storage unit (RBUFF) 421, a reception signal delay processing unit (RDP) 4251, a frequency domain conversion processing unit (FT) 427a, and a first echo canceller unit ( EC1) 6271, a second echo canceller (EC2) 6272, a frequency domain inverse transform processor (IFT) 627c, a frequency domain transform processor (FT) 627e, a frequency domain transform processor (FT) 627g, , A control unit (CONT) 627h, a first signal selection unit (SEL1) 627i, and a second signal selection unit (SEL2) 627j.

  The first echo canceller unit (EC1) 6271 includes a frequency domain adaptive filter unit (FDAF) 6271b, a signal subtraction processing unit 6271d, and a frequency domain adaptive filter control unit (FDCNT) 6271f. Similarly, the second echo canceller unit (EC2) 6272 includes a frequency domain adaptive filter unit (FDAF) 6272b, a signal subtraction processing unit 6272d, and a frequency domain adaptive filter control unit (FDCNT) 6272f.

In fact when the echo length (effective echo length echo suppression) is L, both the two block sizes of the echo canceller unit B, pre-set the 1 ≦ M _D ≦ B-L time shift width M _D satisfying To do. In this embodiment, there are two echo canceller units, a first echo canceller unit (EC1) 6271 and a second echo canceller unit (EC2) 6272. The number of echo canceller units q and the received input signal x [n] The time shift width M _D that is set so that the relationship of D _MAX ≦ q · M _D is established in the maximum fluctuation width D _MAX of the true delay amount D ^* between the echo by the transmission input signal z [n]. The number q of echo cancellers is set according to the value of.

  The operation of each unit of the signal processing device according to the sixth embodiment of the present invention configured as described above will be described with reference to FIG.

  The frequency domain transform processing unit (FT) 627g receives the transmission input signal z [n] output from the A / D converter (A / D) 18 and performs zero padding of 2B-F samples from the beginning. Finally, the transmission input signal z [n] for F samples is arranged and converted to the frequency domain by performing FFT conversion of 2B points, and the frequency spectrum Z [n, ω] of the transmission input signal is converted. Calculate and output.

The first echo canceler unit (EC1) 6271 has a frequency spectrum X ₁ [n, ω] of the received input signal output from the frequency domain transform processing unit (FT) 427a and a second signal selection unit (SEL2) 627j. The frequency spectrum S ′ [n, ω] of the transmission output signal output from the signal and the frequency spectrum Z [n, ω] of the transmission input signal output from the frequency domain transform processing unit (FT) 627g are input. The frequency spectrum Y ′ ₁ [n, ω] of the pseudo echo signal and the power spectrum difference | E ₁ [n, ω] | ² are calculated and output.

Similarly, the second echo canceller unit (EC2) 6272 includes the frequency spectrum X ₂ [n, ω] of the received input signal output from the frequency domain transform processing unit (FT) 427a and the second signal selection unit ( SEL2) The frequency spectrum S ′ [n, ω] of the transmission output signal output from 627j, and the frequency spectrum Z [n, ω] of the transmission input signal output from the frequency domain transform processing unit (FT) 627g, Is input and the frequency spectrum Y ′ ₂ [n, ω] of the pseudo echo signal and the power spectrum difference | E ₂ [n, ω] | ² are calculated and output.

  Here, the first echo canceller unit (EC1) 6271 and the second echo canceller unit (EC2) 6272 use a frequency domain adaptive filter algorithm in which the block adaptive filter is accelerated using a 50% overlap overlap preserving method. The filter length N is equal to the block size B, and the FFT score is 2B.

The frequency domain adaptive filter unit (FDAF) 6271b is a frequency domain adaptive filter composed of a transversal filter having a variable filter coefficient H ₁ [n, ω].

The frequency domain adaptive filter unit (FDAF) 6271b then receives the frequency spectrum X ₁ [n, ω] of the received input signal output from the frequency domain transform processing unit (FT) 427a and the second signal selection unit (SEL2). The frequency spectrum S ′ [n, ω] of the transmission output signal output from 627j and the control information EC1state [n, ω] output from the frequency domain adaptive filter control unit (FDCNT) 6271f are input. When EC1state [n, ω] indicates that the learning is not stopped, the filter coefficient H ₁ [n, ω] is adaptively learned for each frame n and frequency band ω, and the control information EC1state [n, ω] is in the learning stopped state. If it indicates that, adaptive learning is not performed. In this way, the filter coefficient H ₁ [n, ω] is calculated.

Also, the frequency domain adaptive filter unit (FDAF) 6271b has a frequency spectrum X ₁ [n, ω] of the received input signal output from the frequency domain transform processing unit (FT) 427a and a filter coefficient H ₁ [n, ω]. Are used to calculate and output the frequency spectrum Y ′ ₁ [n, ω] of the pseudo echo signal as Y ′ ₁ [n, ω] = H ₁ [n, ω] · X ₁ [n, ω].

Then, the frequency domain adaptive filter unit (FDAF) 6271b performs adaptive learning using a variable step size μ _F1 [n, ω] that controls the update width of the filter coefficient H ₁ [n, ω].

The frequency domain adaptive filter unit (FDAF) 6272b is a processing unit that performs the same operation as the frequency domain adaptive filter unit (FDAF) 6271b, and is a frequency configured by a transversal filter having a variable filter coefficient H ₂ [n, ω]. It is a region adaptive filter.

The frequency domain adaptive filter unit (FDAF) 6272b then receives the frequency spectrum X ₂ [n, ω] of the received input signal output from the frequency domain transform processing unit (FT) 427a and the second signal selection unit (SEL2). The frequency spectrum S ′ [n, ω] of the transmission output signal output from the terminal 627j and the control information EC2state [n, ω] output from the frequency domain adaptive filter control unit (FDCNT) 6272f are input. When EC2state [n, ω] indicates that the learning is not stopped, the filter coefficient H ₂ [n, ω] is adaptively learned for each frame n and frequency band ω, and the control information EC2state [n, ω] is in the learning stopped state. If it indicates that, adaptive learning is not performed. In this way, the filter coefficient H ₂ [n, ω] is calculated.

Similarly, the frequency domain adaptive filter unit (FDAF) 6272b includes the frequency spectrum X ₂ [n, ω] of the received input signal output from the frequency domain transform processing unit (FT) 427a and the filter coefficient H ₂ [n, and the frequency spectrum Y ′ ₂ [n, ω] of the pseudo echo signal is calculated as Y ′ ₂ [n, ω] = H ₂ [n, ω] · X ₂ [n, ω] using To do.

Similarly, the frequency domain adaptive filter unit (FDAF) 6272b performs adaptive learning using a variable step size μ _F2 [n, ω] that controls the update width of the filter coefficient H ₂ [n, ω].

  The frequency domain adaptive filter unit (FDAF) 6271b and the frequency domain adaptive filter unit (FDAF) 6272b include, for example, an adaptive filter based on a linear adaptive algorithm such as an LMS algorithm, an NLMS algorithm, and a sequential least square algorithm, a gradient-limited learning identification method, an adaptation It consists of an adaptive filter based on a nonlinear adaptive algorithm such as a Volterra filter.

The signal subtraction processing unit 6271d includes the frequency spectrum Z [n, ω] of the transmission input signal output from the frequency domain conversion processing unit (FT) 627g and the pseudo echo output from the frequency domain adaptive filter unit (FDAF) 6271b. The frequency spectrum Y ′ ₁ [n, ω] of the signal is input, and | Z [n, ω] | ² − | Y ′ ₁ [n, ω] | ² is the power spectrum difference | E ₁ [n, ω]. | ² is calculated and output.

The signal subtraction processing unit 6272d is a processing unit that performs the same operation as the signal subtraction processing unit 6271d, and the frequency spectrum Z [n, ω] of the transmission input signal output from the frequency domain conversion processing unit (FT) 627g, The frequency spectrum Y ′ ₂ [n, ω] of the pseudo echo signal output from the frequency domain adaptive filter unit (FDAF) 6272b is input, and | Z [n, ω] | ² − | Y ′ ₂ [n, ω ] ^{| 2} of the power spectrum difference _{| E 2 [n, ω]} | 2 calculates and outputs a.

The control unit (CONT) 627h includes the power spectrum difference | E ₁ [n, ω] | ² output from the signal subtraction processing unit 6271d and the power spectrum difference | E ₂ [n, output from the signal subtraction processing unit 6272d. ω] | ² as input, control information ECcontrol [n, ω] indicating which of the first echo canceller (EC1) 6271 and the second echo canceller (EC2) 6272 is selected. Is output.

Specifically, the control unit (CONT) 627h collects all frequencies, and the sum of power spectrum differences | E ₁ [n, ω] | ² of all frequencies is the power spectrum difference of all frequencies | E ₂ [n, If it is smaller than the sum of ω] | ² , the control information ECcontrol [n, ω] is set and output indicating that the first echo canceller (EC1) 6271 has been selected at all frequencies. When the sum of power spectrum differences | E ₂ [n, ω] | ² of all frequencies is smaller than the sum of power spectrum differences | E ₁ [n, ω] | ² of all frequencies, the control information ECcontrol [n , Ω], the fact that the second echo canceller (EC2) 6272 is selected at all frequencies is set and output.

Alternatively, the control information ECcontrol [n, ω] is set and output for each frequency ω. That is, when the power spectrum difference | E ₁ [n, ω] | ² is smaller than the power spectrum difference | E ₂ [n, ω] | ² , the control information ECcontrol [n, ω] includes the frequency ω. The fact that the first echo canceller (EC1) 6271 has been selected is set. When the power spectrum difference | E ₂ [n, ω] | ² is smaller than the power spectrum difference | E ₁ [n, ω] | ² , the control information ECcontrol [n, ω] includes the frequency ω. The fact that the second echo canceller (EC2) 6272 has been selected is set.

The first signal selection unit (SEL1) 627i includes a frequency spectrum Y ′ ₁ [n, ω] of the pseudo echo signal output from the frequency domain adaptive filter unit (FDAF) 6271b, and a frequency domain adaptive filter unit (FDAF) 6272b. The frequency spectrum Y ′ ₂ [n, ω] of the pseudo echo signal output from the control information ECcontrol [n, ω] output from the control unit (CONT) 627h is input, and the control information ECcontrol [n, ω is input. ] Is output as the frequency spectrum Y ′ [n, ω] of the pseudo echo signal output from the selected echo canceller unit.

That is, the first signal selection unit (SEL1) 627i determines that Y ′ [n, ω] = Y when the control information ECcontrol [n, ω] indicates that the first echo canceller unit (EC1) 6271 has been selected. If ' ₁ [n, ω] and the control information ECcontrol [n, ω] indicate that the second echo canceller (EC2) 6272 is selected, Y ′ [n, ω] = Y ′ ₂ [n, ω ].

  The frequency domain inverse transform processing unit (IFT) 627c receives the frequency spectrum Y ′ [n, ω] of the pseudo echo signal output from the first signal selection unit (SEL1) 627i and performs 2B point IFFT conversion. The final F samples are calculated and output as pseudo echo signals y ′ [n] = (y ′ [nF−F + 1], y ′ [nF−F + 2],..., Y ′ [nF]).

  The signal subtraction processing unit 627d includes the transmission input signal z [n] output from the A / D converter (A / D) 18 and the pseudo echo signal y output from the frequency domain inverse conversion processing unit (IFT) 627c. '[N] is used as an input, and the echo component is suppressed by subtracting the pseudo echo signal y ′ [n] from the transmission input signal z [n] for each sample n, and the residual signal after the echo suppression is suppressed. Transmission output signal s ′ [n] = (s ′ [nF−F + 1], s ′ [nF−F + 2],..., S ′ [nF]) = (z [nF−F + 1] −y ′ [nF -F + 1], z [nF-F + 2] -y '[nF-F + 2], ..., z [nF] -y' [nF]). The transmission output signal s ′ [n] is also output to the communication unit (COM) 11 in addition to each unit in the echo suppression processing unit (ECP) 62.

  The frequency domain transform processing unit (FT) 627e uses the transmission output signal s ′ [n] after echo suppression output from the signal subtraction processing unit 627d as an input, and performs zero padding of 2B-F samples from the beginning. Finally, the transmission output signal s ′ [n] for F samples is arranged and converted to the frequency domain by performing FFT conversion of 2B points, and the frequency spectrum S ′ [n, ω of the transmission output signal is converted. ] Is calculated and output.

  The second signal selection unit (SEL2) 627j outputs the frequency spectrum S ′ [n, ω] of the transmission output signal output from the frequency domain transform processing unit (FT) 627e and the control unit (CONT) 627h. The control information ECcontrol [n, ω] is input, and the frequency spectrum S ′ [n, ω] of the transmission output signal is output to the echo canceller selected according to the control information ECcontrol [n, ω].

  That is, when the control information ECcontrol [n, ω] indicates that the first echo canceller (EC1) 6271 has been selected, the second signal selector (SEL2) 627j displays the frequency spectrum S of the transmission output signal. '[N, ω] is output to the frequency domain adaptive filter unit (FDAF) 6271b and the frequency domain adaptive filter control unit (FDCNT) 6271f of the first echo canceller unit (EC1) 6271. In this case, the control information EC2state [n, ω] of the frequency domain adaptive filter control unit (FDCNT) 6272f of the second echo canceller unit (EC2) 6272 may be in a learning stop state.

  When the control information ECcontrol [n, ω] indicates that the second echo canceller (EC2) 6272 has been selected, the second signal selector (SEL2) 627j displays the frequency spectrum S of the transmission output signal. '[N, ω] is output to the frequency domain adaptive filter unit (FDAF) 6272b and the frequency domain adaptive filter control unit (FDCNT) 6272f of the second echo canceller unit (EC2) 6272. In this case, the control information EC1state [n, ω] of the frequency domain adaptive filter control unit (FDCNT) 6271f of the first echo canceller unit (EC1) 6271 may be in a learning stop state.

  Alternatively, the second signal selection unit (SEL2) 627j selects which of the first echo canceller unit (EC1) 6271 and the second echo canceller unit (EC2) 6272 is selected as the control information ECcontrol [n, ω]. Even if the frequency spectrum S ′ [n, ω] of the transmission output signal is output to the first echo canceller unit (EC1) 6271 and the second echo canceller unit (EC2) 6272, respectively. Good.

  A processing flow of the signal processing device according to the sixth embodiment configured as described above will be described with reference to FIG. In addition, about the same operation step as the operation | movement of the signal processing apparatus which concerns on 1st Embodiment demonstrated with reference to FIGS. 7-9, the same code | symbol is attached | subjected and description of the part is abbreviate | omitted.

In FIG. 30, after step S1002, as a buffer process, the reception input signal x _B0 [n] output from the reception signal storage unit (RBUFF) 421 is input, and the reception signal delay processing unit (RDP) 4251 receives the reception input signal x. _B1 [n] is output.

Then, the frequency domain conversion processing unit (FT) 427a converts the reception input signal x _B1 [n] output from the reception signal delay processing unit (RDP) 4251 into the frequency domain by FFT conversion, and the frequency of the reception input signal. A spectrum X ₁ [n, ω] is calculated.

In addition, the frequency domain transform processing unit (FT) 427a outputs the frequency spectrum X ₁ [n−1, ω] of the received input signal one frame before as X ₂ [n−1, ω]. Further, the frequency domain conversion processing unit (FT) 627g converts the transmission input signal z [n] into the frequency domain by FFT conversion, and calculates the frequency spectrum Z [n, ω] of the transmission input signal (step). S6004).

Next, the frequency domain adaptive filter unit (FDAF) 6271b uses the frequency spectrum X ₁ [n, ω] of the received input signal and the filter coefficient H ₁ [n, ω] to generate the frequency spectrum Y ′ ₁ [ n, ω] is calculated. Then, the signal subtraction processing unit 6271d receives the frequency spectrum Y ′ ₁ [n, ω] of the pseudo echo signal and the frequency spectrum Z [n, ω] of the transmission input signal as input, and the power spectrum difference | E ₁ [n, ω ] ² is calculated (step S6009-1).

Similarly, the frequency domain adaptive filter unit (FDAF) 6272b uses the frequency spectrum X ₂ [n, ω] of the received input signal and the filter coefficient H ₂ [n, ω] to generate the frequency spectrum Y ′ ₂ [ n, ω] is calculated. Then, the signal subtraction processing unit 6272d receives the frequency spectrum Y ′ ₂ [n, ω] of the pseudo echo signal and the frequency spectrum Z [n, ω] of the transmission input signal as input, and the power spectrum difference | E ₂ [n, ω ] ² is calculated (step S6009-2).

The control unit (CONT) 627h receives the power spectrum difference | E ₁ [n, ω] | ² and the power spectrum difference | E ₂ [n, ω] | ² as input, and receives control information ECcontrol [n, ω]. Output (step S6007).

  The frequency domain inverse transform processing unit (IFT) 627c performs IFFT conversion on the frequency spectrum Y ′ [n, ω] of the pseudo echo signal to calculate the pseudo echo signal y ′ [n]. Then, the signal subtraction processing unit 627d subtracts the pseudo echo signal y '[n] from the transmission input signal z [n] to calculate the transmission output signal s' [n] (step S6201).

  A frequency domain transform processing unit (FT) 627e transforms the transmission output signal s' [n] into the frequency domain by FFT transform. The second signal selection unit (SEL2) 627j outputs the frequency spectrum S ′ [n, ω] of the transmission output signal to the echo canceller unit selected according to the control information ECcontrol [n, ω]. The frequency domain adaptive filter control unit (FDCNT) 6271f and the frequency domain adaptive filter control unit (FDCNT) 6272f perform control information EC1state [n, ω] and control information EC2state, which are information indicating whether or not learning is stopped, as adaptive filter control processing. [N, ω] are output (step S6202).

Then, the frequency domain adaptive filter unit (FDAF) 6271b receives the control of the control information EC1state [n, ω] output from the frequency domain adaptive filter control unit (FDCNT) 6271f, and receives the filter coefficient H ₁ [n, ω]. To perform adaptive learning.

Similarly, the frequency domain adaptive filter unit (FDAF) 4272b receives the control of the control information EC2state [n, ω] output from the frequency domain adaptive filter control unit (FDCNT) 4272f, and the filter coefficient H ₂ [n, ω ] Is adaptively learned (step S6203).

  However, when the control information ECcontrol [n, ω] is determined collectively for all frequencies, the selected echo canceller unit of the frequency domain adaptive filter unit (FDAF) 6271b or the frequency domain adaptive filter unit (FDAF) 4272b is selected. Only adaptive learning is performed.

  Then, according to the determination processing in step S1006, the processing from step S1002 to step S6203 is performed until the call ends.

  In the above, there are two echo canceller units, the first echo canceller unit (EC1) 6271 and the second echo canceller unit (EC2) 6272, but when the number q of echo canceller units is q> 2, It is configured and operates in the same manner as described above.

  In the above description, the first echo canceller unit (EC1) 6271 and the second echo canceller unit (EC2) 6272 are frequency domain type adaptive systems having no gradient constraint as a high-speed processing method for block processing adaptive filters using FFT transform. Although an example of a filter is shown, it may be configured by a frequency domain adaptive filter with a gradient constraint. The first echo canceler unit (EC1) 6271 and the second echo canceller unit (EC2) 6272 show an example of the overlap preserving method using FFT transform in this embodiment, but the overlap addition method is used. It may be configured or MDCT conversion may be used.

  Due to the operation of the echo suppression processing unit (ECP) 62 of the signal processing apparatus according to the sixth embodiment described above, different delays occur in the first echo canceller unit (EC1) 6271 and the second echo canceller unit (EC2) 6272. The amount of received input signal is input, echo suppression processing is performed, and the result of echo suppression processing is selected by the first signal selection unit (SEL1) 627i according to a predetermined selection criterion, so that the delay amount is not estimated. .

Therefore, it is possible to not made dependent echo suppression processing performance estimation accuracy of delay, stable echo suppression performance can be obtained even when the true delay D ^* of variation, the true delay D ^* variation in Can be robust against. Compared to the modification of the fourth embodiment, the echo suppression processing unit (ECP) 62 instead of the echo suppression processing unit (ECP) 422 reduces the number of times of IFFT conversion, thereby reducing the amount of calculation. be able to.

  Although the echo suppression processing unit (ECP) 62 has two echo canceller units, that is, a first echo canceller unit (EC1) 6271 and a second echo canceller unit (EC2) 6272, three or more The echo canceller may be provided.

  When there are three or more echo canceller units, the input of each echo canceller unit is as described in the operation when the echo suppression processing unit (ECP) 422 has three or more echo canceller units.

  When three or more echo canceller units are included, the control unit (CONT) 627h controls that the echo canceller unit that outputs the power spectrum having the smallest sum of the power spectrum differences of all frequencies is selected at all frequencies. Information ECcontrol [n, ω] is set. Alternatively, the control information ECcontrol [n, ω] is set to indicate that the echo canceller unit that has output the minimum power spectrum for each frequency ω is selected.

By having a plurality of echo canceller units, and receiving the input of each echo canceller unit as a received input signal that dates back in the past, a received input signal corresponding to the true delay amount D ^* in any echo canceller unit it is possible to enter a stable echo suppression performance can be obtained even when the true delay D ^* variation can be made robust to true delay amount D ^* variation.

(Seventh embodiment)
The signal processing device according to the seventh embodiment is different from the signal processing device according to the sixth embodiment in that it does not have the echo suppression processing unit (ECP) 62 but has an echo suppression processing unit (ECP) 72. The other parts are the same. Therefore, the same parts are denoted by the same reference numerals and the description thereof is omitted.

  In addition, the part constituting the echo suppression processing unit (ECP) 72 and the same part as the part constituting the echo suppression processing unit (ECP) 62 according to the sixth embodiment are denoted by the same reference numerals. The description is omitted.

  FIG. 31 is a block diagram showing a configuration of an echo suppression processing unit (ECP) 72 of the signal processing device according to the seventh embodiment of the present invention.

  The echo suppression processing unit (ECP) 72 includes a reception signal storage unit (RBUFF) 421, a reception signal delay processing unit (RDP) 4251, a frequency domain conversion processing unit (FT) 427a, and a first echo canceller unit ( EC1) 7271, second echo canceller (EC2) 7272, frequency domain inverse transform processor (IFT) 627c, frequency domain transform processor (FT) 627e, frequency domain transform processor (FT) 627g, , A control unit (CONT) 727h, a first signal selection unit (SEL1) 627i, and a second signal selection unit (SEL2) 627j.

  The first echo canceller unit (EC1) 7271 includes a frequency domain adaptive filter unit (FDAF) 6271b and a frequency domain adaptive filter control unit (FDCNT) 6271f. Similarly, the second echo canceller unit (EC2) 7272 includes a frequency domain adaptive filter unit (FDAF) 6272b and a frequency domain adaptive filter control unit (FDCNT) 6272f.

Here, the actual time that the echo length (echo valid echo length suppression) was both B block size L, 2 single echo canceller, a 1 ≦ M _D ≦ B-L time shift width M _D satisfying Set in advance. In this embodiment, there are two echo canceller units, a first echo canceller unit (EC1) 7271 and a second echo canceller unit (EC2) 7272. The number of echo canceller units q and the received input signal x [n] The time shift width M _D that is set so that the relationship of D _MAX ≦ q · M _D is established in the maximum fluctuation width D _MAX of the true delay amount D ^* between the echo by the transmission input signal z [n]. The number q of echo cancellers is set according to the value of.

  The operation of each part of the signal processing device according to the seventh embodiment of the present invention configured as described above will be explained with reference to FIG.

The first echo canceller (EC1) 7271 has a frequency spectrum X ₁ [n, ω] of the received input signal output from the frequency domain transform processor (FT) 427a and a second signal selector (SEL2) 627j. The frequency spectrum S ′ [n, ω] of the transmission output signal output from is input, and the frequency spectrum Y ′ ₁ [n, ω] of the pseudo echo signal is calculated and output.

Similarly, the second echo canceller unit (EC2) 7272 includes the frequency spectrum X ₂ [n, ω] of the received input signal output from the frequency domain transform processing unit (FT) 427a and the second signal selection unit ( SEL2) The frequency spectrum S ′ [n, ω] of the transmission output signal output from 627j is input, and the frequency spectrum Y ′ ₂ [n, ω] of the pseudo echo signal is calculated and output.

  Here, the first echo canceller unit (EC1) 7271 and the second echo canceller unit (EC2) 7272 use a frequency domain adaptive filter algorithm in which the block adaptive filter is accelerated using a 50% overlap overlap preserving method. The filter length N is equal to the block size B, and the FFT score is 2B.

The control unit (CONT) 727h includes the frequency spectrum Y ′ ₁ [n, ω] of the pseudo echo signal output from the frequency domain adaptive filter unit (FDAF) 6271b of the first echo canceller unit (EC1) 7271, and the second Frequency spectrum Y ′ ₂ [n, ω] of the pseudo echo signal output from the frequency domain adaptive filter unit (FDAF) 6272b of the echo canceller unit (EC2) 7272 and the frequency domain transform processing unit (FT) 627g. Which echo canceller unit is selected from the first echo canceler unit (EC1) 7271 and the second echo canceler unit (EC2) 7272 with the frequency spectrum Z [n, ω] of the transmitted input signal as an input Control information ECcontrol [n, ω] is output.

More specifically, the control unit (CONT) 727h, the power spectrum of the transmission input signal ^{| Z [n, ω] |} 2 of the power spectrum of the pseudo echo signal than the sum of all frequencies | Y _'1 [n, ω ] | total sum frequency of the ^two is small, and the pseudo echo signal power spectrum _{| Y '1 [n, ω} ] | total sum frequency of ² the power spectrum of the pseudo echo signal _{| Y' 2 [n, ω} ] When it is larger than the sum of all frequencies of | ^2, the fact that the first echo canceller (EC1) 7271 is selected at all frequencies is set in the control information ECcontrol [n, ω].

On the other hand, the power spectrum of the transmission input signal | Z [n, ω] | 2 of the power spectrum of the pseudo echo signal than the sum of all frequencies _{| Y '2 [n, ω} ] | total sum frequency of the ^two is small, And the sum of all the frequencies of the power spectrum | Y ′ ₂ [n, ω] | ² of the pseudo echo signal is larger than the sum of all the frequencies of the power spectrum | Y ′ ₁ [n, ω] | ² of the pseudo echo signal. The control unit (CONT) 727h sets in the control information ECcontrol [n, ω] that the second echo canceller unit (EC2) 7272 has been selected at all frequencies.

Alternatively, the power spectrum | Y ′ ₁ [n, ω] | ² of the pseudo echo signal is smaller than the power spectrum | Z [n, ω] | ^{2 of} the transmission input signal and the power spectrum | Y ′ of the pseudo echo signal. ₁ [n, ω] ^{| 2} is the power spectrum of the pseudo echo signal _{| Y '2 [n, ω} ] | If ² is greater than, the control unit (CONT) 727h, the control information ECcontrol [n, ω] the The fact that the first echo canceller (EC1) 7271 has been selected is set for each frequency ω.

The power spectrum | Y ′ ₂ [n, ω] | ² of the pseudo echo signal is smaller than the power spectrum | Z [n, ω] | ^{2 of} the transmission input signal and the power spectrum | Y ′ of the pseudo echo signal. ₂ [n, omega] ^{| 2} is the power spectrum of the pseudo echo signal _{| Y '1 [n, ω} ] | If ² is greater than, the control unit (CONT) 727h, the control information ECcontrol [n, ω] the The fact that the second echo canceller (EC2) 7272 has been selected is set for each frequency ω.

  A processing flow of the signal processing device according to the seventh embodiment configured as described above will be described with reference to FIG. Note that the same operation steps as those of the signal processing apparatus according to the sixth embodiment described with reference to FIG. 30 are denoted by the same reference numerals, and description thereof is omitted.

In FIG. 32, after step S6004, the frequency domain adaptive filter unit (FDAF) 6271b uses the frequency spectrum X ₁ [n, ω] of the received input signal and the filter coefficient H ₁ [n, ω] as the echo component estimation processing. The frequency spectrum Y ′ ₁ [n, ω] of the pseudo echo signal is calculated (step S7009-1).

Similarly, the frequency domain adaptive filter unit (FDAF) 6272b uses the frequency spectrum X ₂ [n, ω] of the received input signal and the filter coefficient H ₂ [n, ω] as an echo component estimation process, The frequency spectrum Y ′ ₂ [n, ω] is calculated (step S7009-2).

The control unit (CONT) 727h then transmits the frequency spectrum Y ′ ₁ [n, ω] of the pseudo echo signal, the frequency spectrum Y ′ ₂ [n, ω] of the pseudo echo signal, and the frequency spectrum Z of the transmission input signal. [N, ω] as input, control information ECcontrol [n, ω] indicating which one of the first echo canceller (EC1) 7271 and the second echo canceller (EC2) 7272 is selected. ] Is output (step S7007). Then, signal subtraction processing and subsequent steps in step S6201 are performed.

  In the above, there are two echo canceller units, the first echo canceller unit (EC1) 7271 and the second echo canceller unit (EC2) 7272, but when the number q of echo canceller units is q> 2, It is configured and operates in the same manner as described above.

  In the above, the first echo canceller unit (EC1) 7271 and the second echo canceller unit (EC2) 7272 are frequency domain type adaptive systems without gradient constraint as a method for performing high speed processing of block processing adaptive filters using FFT transform. Although an example of a filter is shown, it may be configured by a frequency domain adaptive filter with a gradient constraint.

  The first echo canceler unit (EC1) 7271 and the second echo canceller unit (EC2) 7272 show an example of the overlap preserving method using FFT transform in this embodiment, but the overlap addition method Or MDCT conversion may be used.

By the operation of the signal processing apparatus according to the seventh embodiment described above, reception input signals having different delay amounts are input to the first echo canceller unit (EC1) 7271 and the second echo canceller unit (EC2) 7272. The delay amount is not estimated by performing echo suppression processing and selecting the echo suppression processing result by the first signal selection unit (SEL1) 627i on the basis of the maximum power as long as it is smaller than the transmission input signal. , can not made dependent echo suppression processing performance estimation accuracy of delay, stable echo suppression performance can be obtained even when the true delay D ^* variation, the true delay amount D ^* variation in It can be robust against it. Compared to the sixth embodiment, the echo suppression processing unit (ECP) 72 instead of the echo suppression processing unit (ECP) 62 has less signal subtraction processing in the frequency domain, thereby reducing the amount of calculation. be able to.

  The echo suppression processing unit (ECP) 72 has two echo canceller units, that is, a first echo canceller unit (EC1) 7271 and a second echo canceller unit (EC2) 7272, but three or more. The echo canceller may be provided.

  When there are three or more echo canceller units, the input of each echo canceller unit is as described in the operation when the echo suppression processing unit (ECP) 422 has three or more echo canceller units.

  When three or more echo canceller units are included, the control unit (CONT) 727h outputs a pseudo echo signal in which the sum of all frequencies of the power spectrum is smaller than the sum of all frequencies of the power spectrum of the transmission input signal. If there is an echo canceller, the control information ECcontrol [n, ω] is set to indicate that the echo canceller that outputs the pseudo echo signal of the maximum power spectrum is selected at all frequencies. Or it sets similarly for every frequency (omega).

By having a plurality of echo canceller units, and receiving the input of each echo canceller unit as a received input signal that dates back in the past, a received input signal corresponding to the true delay amount D ^* in any echo canceller unit it is possible to enter a stable echo suppression performance can be obtained even when the true delay D ^* variation can be made robust to true delay amount D ^* variation.

(Eighth embodiment)
The signal processing device according to the eighth embodiment is different from the signal processing device according to the fifth embodiment in that it does not have the echo suppression processing unit (ECP) 52 but has an echo suppression processing unit (ECP) 82. The other parts are the same. Therefore, the same parts are denoted by the same reference numerals and the description thereof is omitted.

  FIG. 33 is a block diagram showing a configuration of an echo suppression processing unit (ECP) 82 of the signal processing device according to the eighth embodiment of the present invention. The echo suppression processing unit (ECP) 82 according to the eighth embodiment has the same part as the echo suppression processing unit (ECP) 52 according to the fifth embodiment. Therefore, the same parts are denoted by the same reference numerals and the description thereof is omitted.

  The echo suppression processing unit (ECP) 82 includes a reception signal storage unit (RBUFF) 521, a reception signal delay processing unit (RDP) 5251, a reception signal delay processing unit (RDP) 5252, and a first echo reduction unit ( ER1) 8271, a second echo reduction unit (ER2) 8272, a frequency domain conversion processing unit (FT) 527b, a transmission power calculation unit (POW) 527d, a frequency domain inverse conversion processing unit (IFT) 527L, , A gain storage unit (GTBL) 327h, an echo suppression gain calculation unit (GCAL) 327i, a signal suppression unit (SS) 327j, and a signal selection unit (SEL) 828.

  FIG. 34 is a block diagram showing a configuration of a Qth echo reduction unit (ERQ) 827Q (where Q is 1 or 2). The Qth echo reduction unit (ERQ) 827Q includes a frequency domain conversion processing unit (FT) 527Qa, a received power calculation unit (POW) 527Qc, an acoustic coupling amount estimation unit (ACLE) 527Qe, and an echo amount estimation unit (ELE). ) 527Qf and a frequency domain control unit (FCNT) 527Qg.

  The operation of each part of the echo suppression processing unit (ECP) 82 of the signal processing device according to the eighth embodiment of the present invention configured as described above will be described with reference to FIGS. 33 and 34. FIG.

The Q-th echo reduction unit (ERQ) 827Q includes a reception input signal x _WQ [n] output from the reception signal delay processing unit (RDP) 525Q, and a transmission output from the transmission power calculation unit (POW) 527d. Using the input signal power spectrum | Z [n, ω] | ² as an input, a smoothed echo amount | Y _QS [n, ω] | ² is calculated and output.

The signal selection unit (SEL) 828 includes the echo amount | Y _1S [n, ω] | ² output from the echo amount estimation unit (ELE) 5271f of the first echo reduction unit (ER1) 8271 and the second echo reduction. Part (ER2) 8272, echo amount | Y _2S [n, ω] | ² output from echo amount estimation unit (ELE) 5272f, and transmission input signal output from transmission power calculation unit (POW) 527d Power spectrum | Z [n, ω] | ² is input, and an echo amount | Y _S [n, ω] | ² is calculated and output.

Specifically, the signal selection unit (SEL) 828 has an echo amount | Y _1S [n, ω] | ² rather than the sum of all frequencies of the power spectrum | Z [n, ω] | ² of the transmission input signal. small sum of all frequencies, and echo amount _{| Y 1S [n, ω]} | entire frequency sum echo amount of _{^{2 | Y 2S [n, ω}} ] | ² If greater than the sum of all frequencies, the total selects the output of the first echo reduction unit (ER1) 8271 at a frequency, the echo amount _{| Y S [n, ω]} | 2 is the echo amount _{| Y 1S [n, ω]} | is output as a ² .

On the other hand, the power spectrum of the transmission input signal ^| Z [n, ω] | echo amount than the sum of all frequencies _{^{2 | Y 2S [n, ω}} ] | total sum frequency of the ^two is small and the echo amount | Y _When the sum of all frequencies of _2S [n, ω] | ² is larger than the sum of all frequencies of echo amount | Y _1S [n, ω] | ² , the signal selection unit (SEL) 828 performs the operation at all frequencies. 2 is selected, and the echo quantity | Y _S [n, ω] | ² is output as the echo quantity | Y _2S [n, ω] | ² .

Or, the power spectrum of the transmission input signal ^| Z [n, ω] | echo amount than _{^{2 | Y 1S [n, ω}} ] | 2 is small and the echo amount _{| Y 1S [n, ω]} | 2 echo If the amount | Y _2S [n, ω] | ² is greater than the amount | Y _2S [n, ω] | ² , the signal selection unit (SEL) 828 selects the output of the first echo reduction unit (ER1) 8271 for each frequency ω, and for each frequency ω. ^{_{Y S [n, ω] |}} | echo amount ^2, the echo amount _{| Y 1S [n, ω]} | is output as a ^2.

On the other hand, the power spectrum of the transmission input signal ^| Z [n, ω] | echo amount than _{^{2 | Y 2S [n, ω}} ] | 2 is small and the echo amount _{| Y 2S [n, ω]} | 2 echo If the quantity | Y _1S [n, ω] | ² is greater than ² , the signal selection unit (SEL) 828 selects the output of the second echo reduction unit (ER2) 8272 for each frequency ω, and for each frequency ω. ^{_{Y S [n, ω] |}} | echo amount ^2, the echo amount _{| Y 2S [n, ω]} | is output as a ^2.

  A processing flow of the signal processing device according to the eighth embodiment configured as described above will be described with reference to FIGS. FIG. 35 is a flowchart showing a processing flow of the entire signal processing apparatus, and FIG. 36 is a flowchart showing a processing flow in the first echo reduction unit (ER1) 8271 and the second echo reduction unit (ER2) 8272. . Note that the same operation steps as those of the signal processing apparatus according to the fifth embodiment described with reference to FIGS. 27 to 28 are denoted by the same reference numerals, and description thereof is omitted.

In FIG. 35, after step S5004, the first echo reduction unit (ER1) 8271 calculates the smoothed echo amount | Y _1S [n, ω] | ² and the second echo reduction unit (ER2) 8272 performs the smoothed echo. The quantity | Y _2S [n, ω] | ² is calculated (step S8005).

Then, the signal selection unit (SEL) 828 includes the echo amount | Y _1S [n, ω] | ² , the echo amount | Y _2S [n, ω] | ^2, and the power spectrum of the transmission input signal | Z [n , Ω] | ² as an input, the echo amount | Y _S [n, ω] | ² is calculated (step S8007). Then, an echo suppression gain calculation process in step S3206 is performed.

  The processes of the first echo reduction unit (ER1) 8271 and the second echo reduction unit (ER2) 8272 shown in FIG. 36 are performed as follows. After performing the echo amount estimation processing in steps S5205-1 and S5205-2, the echo reduction processing is terminated.

  In the above, there are two echo canceller units, the first echo reduction unit (ER1) 8271 and the second echo reduction unit (ER2) 8272, but when the number q of echo canceller units is q> 2, It is configured and operates in the same manner as described above.

  In the above description, the first echo reduction unit (ER1) 8271 and the second echo reduction unit (ER2) 8272 are described as operating as a method of processing for each frequency band by frequency domain conversion by FFT conversion. You may implement | achieve the frequency domain type | mold echo suppression processing, such as the method which puts together the frequency band by FFT conversion, and processes for every frequency band group, the method using MDCT conversion, and band division filters, such as a filter bank.

  By the operation of the signal processing device according to the eighth embodiment described above, the first echo reduction unit (ER1) 8271 and the second echo reduction unit are the same as the operation of the signal processing device according to the fifth embodiment. (ER2) A received input signal having a different delay amount is input to 8272, echo suppression processing is performed, and the signal selection unit (SEL) 828 selects from these echo suppression processing results, so that the delay amount is not estimated. Therefore, the echo suppression processing performance can be made independent of the delay amount estimation accuracy.

  In the above description, the echo suppression processing unit (ECP) 82 has two echo reduction units, that is, a first echo reduction unit (ER1) 8271 and a second echo reduction unit (ER2) 8272. Although mainly, it may have three or more echo reduction units.

  When there are three or more echo reduction units, the input of each echo reduction unit is as described in the operation when the echo suppression processing unit (ECP) 52 has three or more echo reduction units.

  When there are three or more echo reduction units, the signal selection unit (SEL) 828 outputs an echo amount whose sum of all frequencies is smaller than the sum of all frequencies of the power spectrum of the transmission input signal. The output of the echo reduction unit that outputs the maximum echo amount is selected. Alternatively, for each frequency ω, an output of the echo reduction unit that outputs a maximum echo amount that is smaller than the power spectrum of the transmission input signal is selected.

By having a plurality of echo canceller units, and receiving the input of each echo canceller unit as a received input signal that dates back in the past, a received input signal corresponding to the true delay amount D ^* in any echo canceller unit it is possible to enter a stable echo suppression performance can be obtained even when the true delay D ^* variation can be made robust to true delay amount D ^* variation.

(Other embodiments)
Note that the present invention is applied to a call device, and the call device generally refers to a device having a call function. For example, a hands-free call device having a hands-free call function, a telephone, an interphone, a mobile phone, Includes PHS, VoIP software, VoIP system, video conferencing system, etc.

  In each of the above-described embodiments, the echo generated by the reception output signal wrapping around the transmission input signal and acoustically coupled is suppressed from the transmission input signal with reference to the reception input signal. The echo may be suppressed from the received input signal with reference to the transmission output signal, as the signal circulates at the far end (not shown) and is acoustically coupled.

  As an example of signal processing for reducing at least the echo component, an echo canceller and echo reduction have been described. This is because they are general names, and the present invention is not limited to these names. In addition, the present invention can be implemented in a combination of these signal processings without departing from the gist of the present invention.

  The present invention is not limited to the above-described embodiments, and various modifications can be made without departing from the scope of the invention at the stage of implementation. Further, the above embodiments include inventions at various stages, and various inventions can be extracted by appropriately combining a plurality of disclosed constituent elements.

  In addition, for example, even if some structural requirements are deleted from all the structural requirements shown in each embodiment, the problem described in the column of the problem to be solved by the invention can be solved, and the effect described in the effect of the invention Can be obtained as an invention. The present invention is established not only as a device but also as a method and program for realizing the function of the device.

1 is a block diagram showing a configuration of a signal processing device according to a first embodiment of the present invention. The block diagram which shows the structure of the echo suppression process part (ECP) which concerns on the 1st Embodiment of this invention. The block diagram which shows the structure of the echo canceller part (EC) which concerns on the 1st Embodiment of this invention. The figure which shows the range of the delay amount which can suppress the echo component which concerns on the 1st Embodiment of this invention. The figure which shows an example of the impulse response of the echo path which concerns on the 1st Embodiment of this invention. The figure which shows the relationship between the delay amount quantized and the estimated delay amount of the echo component which concerns on the 1st Embodiment of this invention. The flowchart which shows the process of the signal processing apparatus which concerns on the 1st Embodiment of this invention. 5 is a flowchart showing processing related to a delay amount of an echo suppression processing unit (ECP) according to the first embodiment of the present invention. 3 is a flowchart showing processing of an echo canceller unit (EC) according to the first embodiment of the present invention. The block diagram which shows the structure of the echo suppression process part (ECP) of the signal processing apparatus which concerns on the 2nd Embodiment of this invention. The block diagram which shows the structure of the echo canceller part (EC) which concerns on the 2nd Embodiment of this invention. The flowchart which shows the process of the signal processing apparatus which concerns on the 2nd Embodiment of this invention. The flowchart which shows the process regarding the delay amount of the echo suppression process part (ECP) which concerns on the 2nd Embodiment of this invention. The flowchart which shows the process of the echo canceller part (EC) which concerns on the 2nd Embodiment of this invention. The block diagram which shows the structure of the echo suppression process part (ECP) of the signal processing apparatus which concerns on the 3rd Embodiment of this invention. The block diagram which shows the structure of the echo reduction part (ER) which concerns on the 3rd Embodiment of this invention. The figure which shows an example of the window function used with the frequency domain conversion process part (FT) which concerns on the 3rd Embodiment of this invention. The flowchart which shows the process of the signal processing apparatus which concerns on the 3rd Embodiment of this invention. The flowchart which shows the process regarding the delay amount of the echo suppression process part (ECP) which concerns on the 3rd Embodiment of this invention. The flowchart which shows the process of the echo reduction part (ER) which concerns on the 3rd Embodiment of this invention. The block diagram which shows the structure of the echo suppression process part (ECP) of the signal processing apparatus which concerns on the 4th Embodiment of this invention. The flowchart which shows the process of the signal processing apparatus which concerns on the 4th Embodiment of this invention. The flowchart which shows the process of the 1st echo canceller part (EC1) and the 2nd echo canceller part (EC2) which concern on the 4th Embodiment of this invention. The block diagram which shows the structure of the echo suppression process part (ECP) of the signal processing apparatus which concerns on the modification of the 4th Embodiment of this invention. The block diagram which shows the structure of the echo suppression process part (ECP) of the signal processing apparatus which concerns on the 5th Embodiment of this invention. The block diagram which shows the structure of the 1st echo reduction part (ER1) and 2nd echo reduction part (ER2) which concern on the 5th Embodiment of this invention. The flowchart which shows the process of the signal processing apparatus which concerns on the 5th Embodiment of this invention. The flowchart of the process of the 1st echo reduction part (ER1) and the 2nd echo reduction part (ER2) which concern on the 5th Embodiment of this invention. The block diagram which shows the structure of the echo suppression process part (ECP) of the signal processing apparatus which concerns on the 6th Embodiment of this invention. The flowchart which shows the process of the signal processing apparatus which concerns on the 6th Embodiment of this invention. The block diagram which shows the structure of the echo suppression process part (ECP) of the signal processing apparatus which concerns on the 7th Embodiment of this invention. The flowchart which shows the process of the signal processing apparatus which concerns on the 7th Embodiment of this invention. The block diagram which shows the structure of the echo suppression process part (ECP) of the signal processing apparatus which concerns on the 8th Embodiment of this invention. The block diagram which shows the structure of the 1st echo reduction part (ER1) and 2nd echo reduction part (ER2) which concern on the 8th Embodiment of this invention. The flowchart which shows the process of the signal processing apparatus which concerns on the 8th Embodiment of this invention. The flowchart of the process of the 1st echo reduction part (ER1) and the 2nd echo reduction part (ER2) which concern on the 8th Embodiment of this invention.

Explanation of symbols

11 Communication Department (COM)
12, 22, 32, 42, 422, 52, 62, 72, 82 Echo suppression processing unit (ECP)
121, 221, 321, 421, 521 Received signal storage unit (RBUFF)
122 Transmission signal storage unit (SBUFF)
123, 223, 323 Delay amount estimation unit (TDE)
124, 224, 324 Delay quantization unit (DQ)
125, 225, 325, 4251, 4252, 5251, 5252 Received signal delay processor (RDP)
126 Delay amount change detection unit (DCD)
127, 227 Echo canceller (EC)
127a Adaptive filter section (AF)
127b, 227d, 4271d, 4272d, 627d, 6271d, 6272d Signal subtraction processing unit 127c Adaptive filter control unit (CNT)
13 D / A converter (D / A)
18 A / D converter (A / D)
227a, 227e, 327a, 327b, 427a, 527b, 627e, 627g, 4271a, 4271e, 4272a, 4272e, 5271a, 5272a Frequency domain transform processing unit (FT)
227b, 4271b, 4272b, 6271b, 6272b Frequency domain adaptive filter unit (FDAF)
227c, 327L, 4271c, 4272c, 527L, 627c Frequency domain inverse transform processing unit (IFT)
227f, 4271f, 4272f, 6271f, 6272f Frequency domain adaptive filter controller (FDCNT)
327 Echo Reduction Unit (ER)
327c, 5271c, 5272c Received power calculation unit (POW)
327d, 527d Transmission power calculation unit (POW)
327e, 5271e, 5272e Acoustic coupling amount estimation unit (ACLE)
327f, 5271f, 5272f Echo amount estimation unit (ELE)
327g, 5271g, 5272g Frequency domain controller (FCNT)
327h, 5271h, 5272h Gain storage section (GTBL)
327i, 5271i, 5272i Echo suppression gain calculator (GCAL)
327j, 5271j, 5272j Signal suppression unit (SS)
428, 528, 828 Signal selection unit (SEL)
4271, 6271, 7271 First echo canceller (EC1)
4272, 6272, 7272 Second echo canceller (EC2)
5271, 8271 First echo reduction unit (ER1)
5272, 8272 Second echo reduction unit (ER2)
529 Noise estimation unit (NE)
627h, 727h Control unit (CONT)
627i first signal selection unit (SEL1)
627j Second signal selection unit (SEL2)

Claims (22)

First signal storage means for storing a first input signal;
Second signal storage means for storing a second input signal containing an echo component of the first input signal;
A delay amount of an echo component included in the second input signal using the first input signal stored in the first signal storage means and the second input signal stored in the second signal storage means A delay amount estimating means for estimating
Echo suppression means for suppressing at least an echo component included in the second input signal,
The delay amount estimating means quantizes the estimated delay amount with a predetermined quantization step size,
The signal processing apparatus according to claim 1, wherein the echo suppression unit performs echo suppression using the first signal delayed by the delay amount quantized by the delay amount estimation unit as a reference signal.   The signal processing apparatus according to claim 1, wherein the echo suppression unit estimates the echo component using an adaptive filter and suppresses the estimated echo component.   The quantization step size is equal to or less than a filter length of the adaptive filter, or less than a length obtained by subtracting an echo length effective for echo suppression set in advance from the filter length. 3. The signal processing apparatus according to 2.   The signal processing apparatus according to claim 3, wherein the echo suppression unit initializes a filter coefficient of the adaptive filter or a step size of adaptive learning when the quantized delay amount changes.   The echo suppression means converts the first input signal into a frequency domain, estimates an echo component in the frequency domain using the first input signal converted into the frequency domain as a reference signal, and the estimated echo component The signal processing apparatus according to claim 1, wherein:   The quantization step size is equal to or smaller than an effective window length of a window function used in the first frequency domain conversion unit, or smaller than a length obtained by subtracting an overlap portion from the window length. The signal processing apparatus according to claim 5.   The signal processing according to any one of claims 1 to 6, wherein the echo suppression unit initializes an internal state when detecting that the quantized delay amount has changed. apparatus.   The delay amount estimation means estimates a delay amount of the echo component by calculating a cross-correlation between the first input signal and the second input signal. 8. The signal processing device according to any one of items 7. First signal storage means for storing a first input signal;
Second signal storage means for storing a second input signal containing an echo component of the first input signal;
Delay means for outputting a delayed input signal obtained by delaying the first input signal by different delay amounts shifted by a predetermined time interval;
At least two echo estimation means for estimating at least an echo component included in the second input signal by using different delayed input signals output by the delay means as reference signals;
A signal processing apparatus comprising: echo suppression means for generating one echo suppression signal using at least two echo suppression signals generated from at least two echo components estimated by the echo estimation means. Noise estimation means for estimating background noise included in the second input signal;
The echo suppression means uses an echo suppression signal generated from at least two echo components estimated by the echo estimation means, and has a signal power larger than the background noise power estimated by the noise estimation means. The signal processing apparatus according to claim 9, wherein the echo suppression signal is generated at a minimum. First signal storage means for storing a first input signal;
Second signal storage means for storing a second input signal containing an echo component of the first input signal;
Delay means for outputting a delayed input signal obtained by delaying the first input signal by different delay amounts shifted by a predetermined time interval;
At least two echo estimation means for estimating and outputting at least an echo component included in the second input signal by using different delayed input signals output by the delay means as reference signals;
Signal generation means for generating one echo estimation signal from at least two echo estimation signals output from the echo estimation means;
Echo suppression means that suppresses echo using one echo estimation signal generated by the signal generation means.   12. The signal generation means generates the echo estimation signal having a maximum signal power smaller than the power of the second input signal from at least two echo estimation signals. Signal processing equipment.   The signal processing apparatus according to claim 12, wherein at least one of the echo estimation means estimates an echo component using an echo suppression signal corresponding to the echo estimation signal generated by the signal generation means.   The signal processing apparatus according to claim 9, wherein the echo estimation unit estimates the echo component using an adaptive filter.   The predetermined time interval is less than or equal to a filter length of the adaptive filter, or less than or equal to a length obtained by subtracting an echo length effective for echo suppression set in advance from the filter length. 14. The signal processing device according to 14.   The echo estimation means converts the first input signal into a frequency domain, and estimates the echo component in the frequency domain using the converted first input signal as a reference signal. The signal processing device according to claim 13.   The predetermined time interval is equal to or less than an effective window length of a window function used when the first input signal is converted to the frequency domain, or equal to or less than a length obtained by subtracting an overlap portion from the window length. The signal processing apparatus according to claim 16, wherein the signal processing apparatus is provided.   The signal processing apparatus according to claim 15, wherein the predetermined time interval is a multiple of an input frame length of the first input signal.   The first input signal is a reception input signal, and the second input signal is a transmission input signal picked up from a microphone. The signal processing apparatus as described. A signal processing program for suppressing the echo component from a first input signal and a second input signal including the echo component of the first input signal,
A function of estimating a delay amount of the echo component using the first input signal and the second input signal, and quantizing the estimated delay amount;
A signal processing program used in a computer having a function of suppressing the echo component using the first signal delayed by the quantized delay amount as a reference signal. A signal processing program for suppressing the echo component from a first input signal and a second input signal including the echo component of the first input signal,
A plurality of delayed input signals are generated by delaying the first input signal by different delay amounts shifted by a predetermined time interval, and the second input signals are respectively set by using the generated plurality of delayed input signals as reference signals. A function of estimating a plurality of at least echo components included in
A signal processing program used in a computer, having a function of creating an echo suppression signal from the estimated plurality of echo components and suppressing the echo component by the created echo suppression signal. A signal processing program for suppressing the echo component from a first input signal and a second input signal including the echo component of the first input signal,
A plurality of delayed input signals are generated by delaying the first input signal by different delay amounts shifted by a predetermined time interval, and the second input signals are respectively set by using the generated plurality of delayed input signals as reference signals. A function of estimating a plurality of at least echo components included in
A function of generating one echo estimation signal from the plurality of estimated echo components;
A signal processing program used in a computer, having a function of suppressing echo using the generated echo estimation signal.

Images