JP3727800B2 - Echo canceller and voice communication apparatus provided with the echo canceller - Google Patents

Description

[0001]
BACKGROUND OF THE INVENTION
The present invention relates to a portable terminal such as a digital car phone that compresses and transmits audio signals in order to secure a large number of lines, and is a so-called hands-free phone type portable terminal capable of making a call without a telephone. The present invention relates to an echo canceller for suppressing deterioration of a call due to an echo generated in a voice communication apparatus and a voice communication apparatus including the echo canceller.
[0002]
[Prior art]
In recent years, in the field of computers and communications, digital signal processing (DSP) has attracted attention and has been applied to various fields. Since this digital signal processing can easily realize complicated processing such as arbitrary change of characteristic constants and adaptive processing, which was difficult with analog processing, it is a general-purpose technology especially in the fields of audio processing and image processing. It is used as.
[0003]
For example, in a hands-free type telephone used in a car phone or the like, during a hands-free call, the received voice from the speaker may wrap around the microphone and be sent to the other party to generate an acoustic echo. In order to cancel the acoustic echo caused by the wraparound of the received voice and maintain the call quality, an echo canceller generally called an echo canceller is used.
[0004]
In this echo canceller, a pseudo echo signal is generated by the digital signal processing described above, and the echo component contained in the transmission signal is removed by subtracting the pseudo echo signal from the transmission signal of the caller input from the microphone. ing.
[0005]
Further, in the wireless cellular phone device, in order to increase the number of communication channels, a low-bit-rate transmission is performed by a voice encoding device that detects a parameter of a voice signal and sends only the parameter.
[0006]
FIG. 3 is a block diagram showing a configuration of a voice communication apparatus provided with a conventional echo canceller.
[0007]
As shown in FIG. 3, a voice communication device used in a wireless mobile phone device or the like includes a speaker 101, a microphone 102, a D / A converter 103, an A / D converter 104, and a voice decoder (DECODER) 109. A speech encoder (CODER) 110 and an echo canceller 100.
[0008]
The speaker 101 outputs the voice of the far-end speaker (calling partner), and the microphone 102 is used to input the voice of the near-end speaker. The D / A converter 103 performs digital / analog conversion on the far-end speaker signal Xs from the speech decoder 109 and outputs it to the speaker 101. The A / D converter 104 performs analog / digital conversion on the voice input from the microphone 102 to generate a digitized near-end speaker signal Ys. The voice decoder 109 performs a decoding process from a signal of only parameters received by a receiving circuit (not shown) to a voice signal, and generates a far-end speaker signal Xs. The speech encoder 110 detects a parameter of the transmission speech signal from which the echo component has been eliminated by the echo canceller 100, encodes only the parameter, and sends the encoded parameter to a transmission circuit (not shown).
[0009]
In such a configuration, the received signal of only the parameters received by the receiving circuit (not shown) is decoded by the voice decoder 109 and further converted back to the analog call signal by the D / A converter 103. Thereafter, a sound is output from the speaker 106 toward the speaker.
[0010]
On the other hand, the transmission voice of the speaker is collected by the microphone 102 and converted into a transmission signal, and then input to the A / D converter 104, where it is first digitized at a predetermined sampling period. The digitized transmission signal, that is, the near-end speaker signal Ys is input to the speech encoder 110 after the echo component is removed by the echo canceller 100, where it is encoded and illustrated. Supplied to the transmission circuit.
[0011]
Here, the echo canceller 100 includes a digital filter (FIR) 105, a subtracter 106, a filter coefficient update unit (ADP) 107, and a double talk detector (DTD) 108.
[0012]
The digital filter 105 multiplies the far-end speaker signal Xs by a filter coefficient to generate a pseudo echo signal Yss. This pseudo echo signal Yss is supplied to the arithmetic unit 106. The subtractor 106 subtracts the pseudo echo signal Yss from the near-end speaker signal Ys that is the output of the A / D converter 104 to generate a residual echo signal Es that cancels the echo component.
[0013]
Further, the filter coefficient update unit 107 updates the filter coefficient of the digital filter 105. Here, in order to minimize the residual signal that passes through without being erased by the subtractor 106, an arithmetic process for updating a filter coefficient (also referred to as a tap coefficient) is performed. By this filter coefficient update (learning), the transfer function of the digital filter 105 gradually approaches the transfer function of the echo path, and the residual signal level becomes substantially zero when both transfer functions become equal.
[0014]
The update of the filter coefficient by the filter coefficient updating unit 107 cannot be performed during so-called double talk when talking in both directions simultaneously. This is because if the filter coefficient is updated during double talk, the filter coefficient does not converge and the echo component included in the transmitted signal cannot be erased accurately. Therefore, a double talk detector 108 is generally provided to detect the double talk period during communication and prevent the echo canceller 100 from diverging.
[0015]
By the way, the above-described conventional echo canceller 100 is configured to perform processing for each output signal of the AD converter 104, that is, configured in a sample format, and performs the determination processing of the double talk period for each sample. Yes.
[0016]
Also, when both echo cancellation processing and speech encoding / decoding processing are performed by a DSP (Digital Signal Processor), speech coding / decoding processing is generally frame processing, whereas echo cancellation processing is sample processing. Therefore, echo cancellation processing is entered as an interrupt during voice encoding / decoding processing. For this reason, extra processing such as register saving for interrupts is required.
[0017]
[Problems to be solved by the invention]
As described above, since the speech coder / decoder and the echo canceller conventionally have independent configurations (frame processing for the speech coder / decoder and sample processing for the echo canceller), when processing is performed with one DSP Needed interrupt handling. In this case, if hangover processing is added to the double talk result, the sample processing must perform interrupt processing only to change the hangover counter value or determine the counter value. There was a problem that would be very many.
[0018]
In addition, the determination of double talk is usually performed by detecting the rising edge of the sound. However, since the sample processing is performed for each sample, there is a problem that it is difficult to detect the signal state such as the rising edge of the sound. It was.
[0019]
The present invention has been made in view of the above points, and an echo canceller capable of efficiently canceling an echo component without requiring complicated arithmetic processing such as interrupt processing, and voice communication including the echo canceller. An object is to provide an apparatus.
[0020]
[Means for Solving the Problems]
  The echo canceller of the present invention includes a first buffer means for storing far-end speaker signals in units of frames, a second buffer means for storing near-end speaker signals in units of frames, and the first buffer means. Filter means for generating a pseudo echo signal by multiplying the accumulated far-end speaker signal by a filter coefficient, and a pseudo echo generated by the filter means from the near-end speaker signal accumulated in the second buffer means Subtracting means for removing the echo component contained in the near-end speaker signal by subtracting the signal, filter coefficient updating means for updating the filter coefficient based on a predetermined algorithm, and residual echo signal obtained by the subtracting means A third buffer means for storing the frame by frame;A rising edge detecting means for reading the residual echo signal accumulated in the third buffer means, dividing the residual signal into a plurality of blocks, and detecting a rising edge of the near-end speaker signal from a change in power for each block; In response to the accumulation of the residual echo signal for one frame in the third buffer means, the relationship between the residual echo signal and the far-end speaker signal accumulated in the first buffer means is the first. Satisfying the second double talk determination criterion that is easier to detect double talk than the first double talk determination criterion set in response to the rising detection by the rising detection means. Double talk detecting means for detecting double talk and interrupting the update of the filter coefficient by the filter coefficient updating means;It is characterized by comprising.
[0021]
  According to such a configuration,Even when the voice rise is difficult to detect double talk, if the rise detection means detects the voice rise, the double talk detection means detects the double talk using a determination criterion that makes it easier to determine the double talk. Therefore, it is possible to make it difficult to erroneously determine single talk.
[0023]
DETAILED DESCRIPTION OF THE INVENTION
Hereinafter, an embodiment of the present invention will be described with reference to the drawings.
[0024]
FIG. 1 is a block diagram showing a configuration of a voice communication apparatus provided with an echo canceller of the present invention.
[0025]
The voice communication device according to the present embodiment is applied to a wireless mobile phone device such as a car phone, for example. As shown in FIG. 1, this device includes a speaker 201, a microphone 202, and a D / A converter 203. , An A / D converter 204, a speech decoder (DECODER) 209, a speech encoder (CODER) 210, and an echo canceller 200.
[0026]
The speaker 201 outputs the voice of the far-end speaker (calling partner), and the microphone 202 is used to input the voice of the near-end speaker. The D / A converter 203 performs digital / analog conversion on the far-end speaker signal Xs from the speech decoder 209 and outputs the converted signal to the speaker 201. The A / D converter 204 performs analog / digital conversion on the voice input from the microphone 202 to generate a digitized near-end speaker signal Ys. The speech decoder 209 performs a decoding process from a signal having only parameters received by a receiving circuit (not shown) to a speech signal, and generates a time-series far-end speaker signal Xs. The speech encoder 210 detects a parameter of the transmission speech signal from which the echo component is eliminated by the echo canceller 200, and encodes only the parameter.
[0027]
Here, the echo canceller 200 includes a digital filter (FIR) 205, a subtracter 206, a filter coefficient update unit (ADP) 207, and a double talk detector (DTD) 208. Further, in the present embodiment, in order to perform the echo cancellation processing in the echo canceller 200 in a frame format, the frame buffer (BUFF) 211 to 214 is provided, and the address generator 215 and the rising detector 216 are provided.
[0028]
The digital filter 205 multiplies the signal sequence of the far-end speaker signal Xs by a filter coefficient to generate a pseudo echo signal Yss. This pseudo echo signal Yss is supplied to the calculator 206. The subtracter 206 subtracts the pseudo echo signal Yss from the near-end speaker signal Ys that is the output of the A / D converter 204 to cancel the echo component, and generates a residual echo signal Es that is the remaining component.
[0029]
The filter coefficient update unit 207 updates the filter coefficient of the digital filter 205 based on a predetermined algorithm. The filter coefficient updating unit 207 performs update processing of a filter coefficient (also referred to as a tap coefficient) for minimizing a residual signal that passes through without being erased by the subtractor 206.
[0030]
As a filter coefficient update algorithm, for example, a learning identification method (NLMS) obtained by normalizing the least mean square method (LMS) is used. This algorithm based on the learning identification method has the advantage that it requires a relatively small amount of calculation and exhibits good characteristics. By this filter coefficient update (learning), the transfer function of the digital filter 205 gradually approaches the transfer function of the echo path, and the residual signal level becomes substantially zero when both transfer functions become equal.
[0031]
The double talk detector 208 detects the double talk period and prevents the echo canceller 200 from diverging.
[0032]
Each frame buffer (BUFF) 211 to 214 is for accumulating each signal for a predetermined time in units of frames. The length of the frame corresponds to the signal processing unit in speech decoder 209 and speech encoder 210.
[0033]
The frame buffer 211 is a frame buffer for near-end speaker signals. The frame buffer 211 is provided in the preceding stage of the subtracter 206, and temporarily stores the near-end speaker signal Ys, which is the output of the A / D converter 204, for one frame. The frame buffer 212 is a frame buffer for pseudo echo signals. The frame buffer 212 is provided before the subtractor 206, and temporarily stores the pseudo echo signal Yss, which is the output of the digital filter 205, for one frame. The frame buffer 213 is a frame buffer for residual echo signals. The frame buffer 213 is provided at the subsequent stage of the subtracter 206, and temporarily stores the residual echo signal Es, which is the output of the subtracter 206, for one frame. The frame buffer 214 is a frame buffer for far-end speaker signals. The frame buffer 214 is provided at the subsequent stage of the speech decoder 209, and temporarily stores the far-end speaker signal Xs, which is the output of the speech decoder 209, for one frame.
[0034]
The address generator 215 automatically sets the address pointers for the near-end speaker signal frame buffer 211, the pseudo echo signal frame buffer 212, and the residual echo signal frame buffer 213 that are used for the subtraction process in the subtractor 206. It is for incrementing. According to the address pointer incremented by the address generator 215, the near-end speaker signal Ys and the pseudo echo signal Yss are fetched by one frame into predetermined areas of the frame buffer 211 and the frame buffer 212, respectively, and obtained as a subtraction result. The residual echo signal Es is taken into a predetermined area of the output buffer of the subtracter 206, that is, the frame buffer 213.
[0035]
The rising detector 216 detects the rising edge of the near-end speaker signal Ys. The rise detector 216 divides a buffer area used for rise detection as described later, calculates a parameter of a signal used for double talk detection for each divided area, and arranges each parameter in ascending order. It is determined that the voice signal is rising. The result of this rising edge detection is reflected in the double talk detector 208.
[0036]
Next, the processing operation of this apparatus will be described.
[0037]
A received signal having only parameters received by a receiving circuit (not shown) is decoded by a speech decoder 209 and converted into a time-series far-end speaker signal Xs. The far-end speaker signal Xs is returned to an analog call signal by the D / A converter 203 and then given to the speaker 206. As a result, the voice of the far-end speaker (of the other party) is output from the speaker 201 toward the near-end speaker.
[0038]
On the other hand, the voice of the near-end speaker is collected by the microphone 202 and converted into a transmitted voice signal, and then input to the A / D converter 204, where it is first digitized at a predetermined sampling period. The digitized transmission signal, that is, the near-end speaker signal Ys is input to the speech encoder 210 after the echo component is removed by the echo canceller 200, where it is encoded and illustrated. Supplied to the transmission circuit.
[0039]
Here, the echo canceller 200 is configured as a frame-type echo canceller, and each frame buffer receives each signal necessary for echo cancellation processing in units of frames that are processing units of the speech decoder 209 and speech encoder 210. The processing is performed while accumulating in 214 to 211.
[0040]
That is, the near-end speaker signal Ys which is the output of the A / D converter 204 and the pseudo echo signal Yss which is the output of the digital filter 205 are stored in the frame buffer 211 and the frame buffer 212 for one frame, and then subtracted. Subtractor 206 performs the subtraction process. By the subtraction processing of the subtracter 206, the echo component included in the near-end speaker signal Ys is removed. The residual echo signal Es obtained as a subtraction result is stored in the frame buffer 213. The residual echo signal Es is given to the filter coefficient updating unit 207, where the filter coefficient update process (learning process) is performed based on a predetermined algorithm.
[0041]
Such echo cancel processing is repeatedly performed. In this case, unlike the conventional sample processing, since the audio signals obtained in time series can be processed in units of frames, there is an advantage that the DSP calculation amount can be greatly reduced.
[0042]
Further, as already described, since there is a case where two-way conversations occur simultaneously during a call, so-called double talk occurs. If the filter coefficient is updated during the double talk period, the filter coefficient does not converge and the echo cancellation process cannot be performed accurately. Therefore, when double talk is detected by the double talk detector 208, it is necessary to interrupt the update of the filter coefficient.
[0043]
Here, the double-talk detector 208 is used for double-talk detection by utilizing the characteristics of the signals accumulated in the buffers described above. Here, the buffer area of the signal used for double-talk detection is divided into two or more, and the parameters used for double-talk detection are calculated for each of the divided areas to determine whether or not the voice rises. In this case, the rising detection result is reflected in the determination of double talk.
[0044]
For example, the frame power characteristic (power value or peak value) of the far-end speaker signal Xs stored in the far-end speaker signal frame buffer 214 and the residual echo signal stored in the residual echo signal frame buffer 213 When performing double talk determination based on the frame power characteristic (power value or peak value) of Es, the determination is performed using the following equation.
[0045]
logX + th <logE (1)
Here, logX is a logarithmic value of the frame power of the far-end speaker signal Xs, and logE is a logarithmic value of the frame power of the residual echo signal Es. Further, th is a variable threshold value, and this is for compensating for the change since the residual echo signal Es becomes smaller as the filter coefficient converges.
[0046]
In the above equation (1), the area of the residual echo signal Es, that is, the residual echo signal frame buffer 213 is divided into four, and the powers of signals in the four divided areas are respectively Ep1, Ep2, Ep3, When Ep4 is set, the values from Ep1 to Ep4 are
Ep1 <Ep2 <Ep3 <Ep4 (2)
If it is, it is determined that the voice signal of the near-end speaker signal Ys is rising, and the above equation (1) is changed as follows.
[0047]
logX + th<logE + th2 (3)
Th2 is a positive variable threshold value, and by adding this variable threshold value th2, it is possible to prevent erroneous determination as single talk when the voice rising power is relatively small at the value of E. That is, it is difficult to detect double talk because the power is relatively low when the voice rises. Therefore, by adding a variable threshold th2 to the residual echo signal Es, it is easy to detect double talk. This is a method peculiar to frame processing and cannot be performed by conventional sampling processing.
[0048]
In the above (3), the threshold value th2 is added to the expression on the right side, but the same applies to subtracting the threshold value th2 from the expression on the left side as shown in the following expression (4). In short, it is only necessary to perform an operation considering that the power of the audio signal is low when the audio rises.
[0049]
logX + th-th2<logE (4)
The rising detector 216 shown in FIG. 1 divides the residual echo signal frame buffer 213 into a plurality of areas as described above, and compares the signal power in each area, thereby When the condition is satisfied, it is determined that the voice signal of the near-end speaker signal Ys is rising, and the result is notified to the double talk detector 208. In the double talk detector 208, the power of each signal of the far-end speaker signal Xs stored in the far-end speaker signal frame buffer 214 and the residual echo signal Es stored in the residual echo signal frame buffer 213. By comparing the characteristics, it is determined that the double talk is being performed when the condition of the expression (1) is satisfied. At this time, when it is detected from the rising detector 216 that the voice is rising, the double talk detector 208 replaces the equation (1) with the equation (3) or (4) to determine the double talk. I do.
[0050]
Thus, by reflecting the detection result of the voice rise in the double talk detection processing, the double talk period can be accurately detected even when the voice rises (when the signal power is low). As a result, the filter coefficient update operation in the filter coefficient update unit 207 can be controlled to correctly perform the echo cancellation process.
[0051]
In the above embodiment, the double-talk period is set by detecting the power characteristic of the residual echo signal Es obtained by subtracting the pseudo echo signal Yss from the far-end speaker signal Xs and the near-end speaker signal Ys accumulated for a predetermined time. However, as another method, it is also possible to detect the power characteristics of the near-end speaker signal Ys and far-end speaker signal Xs accumulated for a predetermined time and to detect the power characteristics of those speech signals as in the above case. It is possible to detect the talk period.
[0052]
The digital filter 205 is a frame type, and the far-end speaker signal Xs for one frame stored in the output buffer in the speech decoder 209 and the filter coefficient updated by the filter coefficient updating unit 207 Is used to generate a pseudo echo signal Yss for the length of the frame by processing at once for the length of the frame. For this reason, the input and output buffer addresses of the digital filter 205 need only be repeatedly added and subtracted, and the amount of calculation, particularly address setting processing, can be reduced accordingly.
[0053]
Further, the order of arrangement of the buffer areas of the filter coefficients in the digital filter 205 is arranged from the lowest address to the higher coefficient order.
[0054]
For example, if the top address of the coefficient area is 10h, and the 10th address of the coefficient area is the highest value of the coefficient, that is, the filter order is 20th, the 20th order is 10th. Arrange the data. Then, 19th order data is assigned to the 11h address, 18th order data to the 12h address, 17th order data to the 13h address, and so on.
[0055]
With this arrangement, when frame processing is performed, both the filter coefficient and the buffer area for the far-end speaker signal Xs are incremented by one step for the filter order, and then the above two buffers are set to the filter order. After returning to the original amount, only the far-end speaker signal buffer area may be incremented by one step. As a result, the address pointer of the two-input buffer area of the digital filter 205 operates in substantially the same manner, and the address generator for that purpose can be simplified.
[0056]
In the above-described embodiment, the frame buffers 211 to 214 are provided in the echo canceller 200. However, even if these frame buffers 211 to 214 are not newly provided, as described below, the audio decoder 209 and the audio buffers Since the encoder 210 already has a buffer for frame processing, this buffer can also be used.
[0057]
FIG. 2 shows a general configuration of speech encoder 210.
[0058]
FIG. 2 mainly shows a coding part of CELP (code excited linear prediction). The DC signal is removed from the input signal to the coder through a high pass filter (HPF) 301. The output of the high-pass filter 301 is obtained by detecting a frequency envelope characteristic parameter by a short-term correlation detector 302 (for example, constituted by an LPC or LSP parameter detector), and generally a quantizing unit that quantizes the detected parameter. To 306.
[0059]
Further, the pitch component is detected by the long-term correlation detector 303 (for example, constituted by a pitch detector), and the quantized value is passed to the multiplexing unit 306. Further, an excitation signal that is input to the filter coefficients obtained by the long-term correlator 303 and the short-term correlator 302 is obtained by an excitation signal detector 304 (for example, constituted by a noise code book detector), and the noise code book The number is passed to the multiplexing unit 306, and the signal bundled by the multiplexing unit 306 is transmitted.
[0060]
The above-described processing of the speech encoder 210, particularly the short-term correlator, requires a certain amount of speech data, and thus the buffer 307 is necessary. If this buffer 307 is used as the frame buffers 211 to 214 for the echo canceller 200 described above, a voice communication apparatus having a smaller number of steps than the prior art can be realized without newly adding more static memory areas than necessary. be able to.
[0061]
The same applies to the audio decoder 209, and a buffer used for the decoding process can be used.
[0062]
In this way, in a voice communication device that combines a voice encoding / decoding device and an echo canceller, each signal can be obtained by making echo cancellation processing in the echo canceller into a frame format that matches the voice encoding / decoding device. Therefore, the echo can be efficiently erased without requiring complicated arithmetic processing such as interrupt processing.
[0063]
Furthermore, by using a buffer included in the speech encoding / decoding device, it is possible to reduce the memory capacity required for echo cancellation processing in the echo canceller, and to perform operations with fewer steps than in the past.
[0064]
【The invention's effect】
  As described above, according to the present invention,Even when the voice rises easily misidentified as single talk, when the voice rise is detected, double talk detection is performed using a criterion that makes it easier to judge double talk. Can do.
[Brief description of the drawings]
FIG. 1 is a block diagram showing a configuration of a voice communication apparatus including an echo canceller according to the present invention.
FIG. 2 is a block diagram showing a configuration of a speech encoder provided in the speech communication apparatus.
FIG. 3 is a block diagram showing a configuration of a voice communication apparatus provided with a conventional echo canceller.
[Explanation of symbols]
201 ... speaker
202 ... Microphone
203 ... D / A converter
204 ... A / D converter
205: Digital filter
206 ... subtractor
207 ... Filter coefficient update unit
208 ... Double talk detector
209 ... Voice decoder
210: Speech encoder
211 ... Near-end speaker signal frame buffer
212 ... Pseudo echo signal frame buffer
213 ... Frame buffer for residual echo signal
214 ... Frame buffer for far-end speaker signal
215 ... Address generator
216 ... Rising detector
301 ... High-pass filter
302 ... Short-term correlation detector
303 ... Long-term correlation detector
304 ... Excitation signal detector
305 ... Gain detector
306: Multiplexer
307 ... Buffer

Claims (5)

First buffer means for storing far-end speaker signals in units of frames;
Second buffer means for storing the near-end speaker signal in units of frames;
Filter means for generating a pseudo echo signal by multiplying the far-end speaker signal accumulated in the first buffer means by a filter coefficient;
Subtracting means for removing an echo component contained in the near-end speaker signal by subtracting the pseudo echo signal generated by the filter means from the near-end speaker signal stored in the second buffer means;
Filter coefficient updating means for updating the filter coefficient based on a predetermined algorithm;
Third buffer means for storing the residual echo signal obtained by the subtracting means in units of frames;
A rising edge detecting means for reading the residual echo signal accumulated in the third buffer means, dividing the residual signal into a plurality of blocks, and detecting a rising edge of the near-end speaker signal from a change in power for each block;
In response to the accumulation of the residual echo signal for one frame in the third buffer means, the relationship between the residual echo signal and the far-end speaker signal accumulated in the first buffer means is the first. Satisfying the second double talk determination criterion that is easier to detect double talk than the first double talk determination criterion set in response to the rising detection by the rising detection means. An echo canceller comprising: a double-talk detecting unit that detects double-talk and interrupts updating of the filter coefficient by the filter coefficient updating unit .   The subtracting means performs a subtraction process by increasing the address pointer of each buffer area for storing the near-end speaker signal, the pseudo echo signal, and the residual echo signal by a certain amount for each process. The echo canceller according to claim 1.   2. The echo canceller according to claim 1, wherein the filter means has a coefficient buffer area, and in the coefficient buffer area, filter coefficients of higher order are arranged in order from the lowest address. A speech coder / decoder for decoding and receiving a far-end speaker signal, encoding and transmitting a near-end speaker signal, and removing an echo component included in the near-end speaker signal In an audio communication device equipped with an echo canceller for
The echo canceller is
First buffer means for storing far-end speaker signals in units of frames;
Second buffer means for storing the near-end speaker signal in units of frames;
Filter means for generating a pseudo echo signal by multiplying the far-end speaker signal accumulated in the first buffer means by a filter coefficient;
Subtracting means for removing an echo component contained in the near-end speaker signal by subtracting the pseudo echo signal generated by the filter means from the near-end speaker signal stored in the second buffer means;
Filter coefficient updating means for updating the filter coefficient based on a predetermined algorithm;
Third buffer means for storing the residual echo signal obtained by the subtracting means in units of frames;
A rising edge detecting means for reading the residual echo signal accumulated in the third buffer means, dividing the residual signal into a plurality of blocks, and detecting a rising edge of the near-end speaker signal from a change in power for each block;
In response to the accumulation of the residual echo signal for one frame in the third buffer means, the relationship between the residual echo signal and the far-end speaker signal accumulated in the first buffer means is the first. Or a second double talk determination criterion that is easier to determine as a double talk than the first double talk determination criterion set in response to detection of a rising edge by the rising detection means. A voice communication apparatus comprising: a double talk detecting means for detecting double talk and interrupting the updating of the filter coefficient by the filter coefficient updating means . The buffer means for storing the far-end speaker signal, the near-end speaker signal and the residual echo signal in units of frames, and receiving and decoding the far-end speaker signal stored in the buffer means, A speech encoder for encoding and transmitting the near-end speaker signal stored in the buffer means, and an echo canceller for removing an echo component included in the near-end speaker signal In a voice communication device,
The echo canceller is
Filter means for generating a pseudo echo signal by multiplying the far-end speaker signal accumulated in the buffer means by a filter coefficient;
Subtracting means for removing an echo component contained in the near-end speaker signal by subtracting the pseudo echo signal generated by the filter means from the near-end speaker signal stored in the buffer means;
Filter coefficient updating means for updating the filter coefficient based on a predetermined algorithm;
Read the residual echo signal accumulated in the buffer means, dividing the residual echo signal into a plurality of blocks, rising detection means for detecting the rising edge of the near-end speaker signal from the change in power for each block;
In response to the accumulation of the residual echo signal for one frame in the buffer means, the relationship between the residual echo signal and the far-end speaker signal is raised by the first double talk determination criterion or the rising edge detecting means. Double-talk is detected by satisfying a second double-talk criterion that is easier to determine as double-talk than the first double-talk criterion set according to the detection, and the filter coefficient updating means performs the above-mentioned A voice communication apparatus comprising: a double talk detecting means for interrupting the update of the filter coefficient .

Images