JP2011055494A - Echo canceller - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To reduce feeling of echo in the subsequent conversation even when a tone signal is input. <P>SOLUTION: The echo canceler uses an adaptive filter for generation of a pseudo echo and removes an echo using a hybrid circuit. The echo canceller includes a tone type determining means for determining whether a remote end input signal is a predetermined type of tone signal, and a band stop filter provided at an input stage of the adaptive filter for removing a frequency component of the determined predetermined type of tone signal from the remote end input signal to the adaptive filter when the tone type determining means determines that the remote end input signal is the predetermined type of tone signal. <P>COPYRIGHT: (C)2011,JPO&INPIT

Description

  The present invention relates to an echo canceller and can be applied to, for example, an echo canceller provided in a VoIP terminal accommodating a telephone terminal.

  An echo canceller that removes an echo component in a hybrid circuit is generally provided to prevent a received signal from flowing through a path of a transmission signal through the hybrid circuit and becoming an echo component to reduce voice quality. . By the way, a reception signal that can be an echo component is generally an audio signal, but a tone signal may be a reception signal.

  Conventionally, there is an echo canceller described in Patent Document 1 in consideration of a case where a tone signal becomes a received signal. The configuration of the echo canceller described in Patent Document 1 will be clarified in the description of the problem.

JP 2005-110307 A

  The echo canceller described in Patent Document 1 has the following problems 1 to 4. The echo canceller described in Patent Document 1 is hereinafter referred to as a conventional technique.

(Problem 1)
Since it is not known that the tone signal is input to the echo canceller (or the adaptive filter) only after the coefficient of the adaptive filter is destroyed by the tone signal, in many cases, it is too late even if some measures are taken, Even if re-convergence is attempted, the coefficient value is inappropriate as the initial state of convergence, so it takes time until the performance can be exhibited again.

  The prior art basically has a mechanism that can detect that the signal is a tone signal after the adaptive filter converges on the tone component. That is, even before a tone signal is input to the adaptive filter, even if a coefficient having a suitable frequency characteristic in a wide band can be estimated by voice or the like, the filter coefficient once has tone characteristics. In order to determine whether or not the coefficient matches the frequency characteristic of the tone after being destroyed, it is often too late to use for various processes following the determination, for example, whether or not the coefficient of the adaptive filter can be updated. Even if the echo canceller is in the initial state, or when reconvergence is executed once in the converged state, the coefficient of the adaptive filter erroneously converged by the tone signal is appropriate as the initial value for reconvergence. Therefore, reconvergence takes time, and during that time, the echo component does not disappear, so the call quality deteriorates.

(Problem 2)
When the initial delay is small, the desired operation is not performed.

  According to the prior art, when the echo canceller estimates an echo path using a tone signal as a reference signal, the adaptive filter coefficient register of the echo canceller converges to the input tone frequency. However, when the initial delay of the echo path is small, the conventional technique often does not exert an effect. This is because, when the initial delay of the echo path is small, the coefficient register actually does not converge to the input tone frequency in many cases.

  Hereinafter, Problem 2 of the related art will be described with reference to FIGS. First, a case where the conventional technology can be applied will be shown, and then an actual example of a case where the conventional technology cannot be applied will be shown.

  FIG. 2A shows an echo path with an initial delay of 50 samples. As shown in FIG. 2C, this echo path is an echo path having a substantially flat frequency characteristic from 0 to 8 kHz. FIG. 2B shows a result obtained by inputting a tone (a dial tone of 400 Hz in this example) to the echo canceller and converging the coefficient register. FIG. 2C shows both the frequency characteristic of the echo path and the frequency characteristic of the coefficient register after convergence. As can be seen from FIG. 2B, the adaptive filter coefficients of the echo canceller have converged to a state that is very similar to the input tone in terms of waveform, and from FIG. It can be seen that only the tone frequency portion has a large component.

  On the other hand, FIGS. 3A to 3C show a state in which the echo canceller is operated in an echo path having an initial delay of 10 samples. FIG. 3A shows an echo path with an initial delay of 10 samples, and the frequency characteristic is an echo path that is substantially flat from 0 to 8 kHz as shown in FIG. Similarly to the above, FIG. 3B shows the result of convergence of the coefficient register by inputting a tone (400 Hz dial tone) to the echo canceller. In FIG. 3B, the adaptive filter coefficient of the echo canceller is different from that of the input tone, and it can be seen from FIG. 3C that is the frequency characteristic that there is no peak at the input tone of 400 Hz. More specifically, in FIG. 3C, the maximum frequency characteristic is 0 Hz.

  However, both cases shown in FIG. 2 and FIG. 3 exhibit sufficient performance with respect to echo cancellation for tone input. In the example of description, both have ERLE (power ratio of a received signal and a residual) of about 35 dB. In other words, despite the fact that the characteristics of the echo path are not sufficiently estimated, there is a state where the echo canceling performance is exhibited only for the input frequency. Such a phenomenon occurs frequently when the initial delay of the echo path is smaller than the tone period. Specifically, it always occurs when the echo canceller and the hybrid circuit as an echo generation source are close to each other, for example, in the same apparatus. Unfortunately, in a VoIP terminal device or the like, it is normal that the hybrid circuit and the echo canceller are arranged in the same device. Therefore, in such a case, the frequency characteristic of the coefficient register does not have an input tone, so the tone characteristic of the coefficient cannot be detected, and the conventional technique does not exhibit an effect.

(Problem 3)
Echo removal performance is severely degraded by double talk immediately after a tone is input to the echo canceller.

  Here, double talk will be explained. A state where both the calling speaker and the called speaker are producing sound is called double talk. In the echo canceller, a transmission signal and a reception signal are input simultaneously. It is in a state. As described in the prior art, in echo cancellers, in the case of double talk, control for stopping coefficient update of the adaptive filter is performed, so that ERLE is often monitored to detect double talk. In the prior art, as shown in the above-described problem 1, the double talk determination is performed using the residual based on the result of using the adaptive filter that has once erroneously converged on the tone signal. Since the filter residual of misconvergence is used, correct double talk cannot be determined, and in the prior art, a step gain for controlling the coefficient update speed is set large for a certain period after tone convergence. As a result, the adaptive filter coefficient is applied while the coefficient update followability is most agile for a certain period after tone convergence. As a result, the sound quality is very deteriorated because of inappropriate cancellation. In addition, even when double talk does not occur, the refocusing of the adaptive filter is the same as the case where the error signal converged to the tone signal is started as an initial state. It takes a long time before it can be removed. Naturally, the speaker hears an echo for a long time until re-convergence, and the call quality deteriorates.

(Problem 4)
When the prior art is used as it is, the echo removability deteriorates at the time of telephone transfer such as extension transmission. Hereinafter, the problem at the time of extension transfer, which is the problem 4 of the related art, will be described with reference to FIG.

  Referring to FIG. 4, a case will be described in which a call is made from the telephone 100 to the telephone 109, a call is made for a while, and then the extension is transferred to the telephone 110. As in the prior art, the echo canceller 104 uses the output of the adder 119 and the input of the reception input terminal 103 to estimate the echo path using the hybrid circuit 108 as an echo path, and sets the coefficient of the adaptive filter 118 to that of the hybrid circuit 108. Estimate to be equal to the impulse response. As a result of such estimation, the speaker signal from the telephone 100 passes through the hybrid circuit 101 and is digitally converted by the analog / digital converter (A / D converter) 102 to become a digital signal. It is converted into an analog signal by a converter (D / A converter) 105, passes through a switch 106, is converted into two lines by a hybrid circuit 108, and then reaches a telephone set 109. A part of the speaker signal to the telephone 109 is reflected by the hybrid circuit 108 to become an echo signal y1, which is converted into a digital signal by the A / D converter 113 via the switch 112 and then input to the adder 119. Is done. The pseudo echo signal y ′ output from the adaptive filter (ADF) 118 is input to the adder 119. If the adaptive filter 118 has converged, y′≈y1, so echo (echo signal y ′). Will be countered.

  A known algorithm such as NLMS is used to update the coefficient of the adaptive filter 118. Here, how the coefficient of the NLMS algorithm is updated will be briefly explained.

Assuming that the voice signal input from the far-end telephone 100 to the reception input terminal is x (n) and the filter coefficient of the adaptive filter 118 is h _k (n), the filter coefficient h _k (n) is expressed by Equation (1) and (2) ) Update to follow the formula.

When the denominator of the second term on the right side of equation (1) is 0, the coefficient update amount is set to 0, or the coefficient update is stopped. The filter coefficient h _k (n) represents the tap coefficient of the kth adaptive filter at the nth sample time. Α is a step gain, a constant that determines the convergence speed of the adaptive filter 118, and 0 <α <2. If α is large, the convergence speed is fast, but the fluctuation width of the steady characteristic is large, and the influence of noise is also large. On the other hand, if α is small, the convergence speed is slow, but the fluctuation width of the steady characteristics is small, and the influence of noise is also small. According to the prior art, it is shown that the step gain is set to a small value after the adaptive filter 118 has sufficiently converged.

  Now, at the time of extension transfer, a telephone call is once made between the telephones 100 and 109, and the coefficient of the adaptive filter 118 is also estimated by the hybrid circuit 108 and converged to some extent. Here, if the adaptive filter 118 is completely converged, the echo removal residual e (n) = 0 in the equation (2) becomes 0, and thus the coefficient update is the same as when the coefficient update is stopped.

  Thereafter, the case where a call is transferred from the telephone 109 to the telephone 110 by a private branch exchange or the like will be described below in four cases. Here, when transfer occurs, the terminal of the switch 106 is closed to a2, and the terminal of the switch 112 is closed to b2. The signal output from the D / A converter 105 passes through the terminal a2 of the switch 106, is partly output to the telephone 110 via the hybrid circuit 111, and part is reflected by the hybrid circuit 111 to be echo signal y2. The signal is input to the A / D converter 113 via the terminal b2 of the switch 112 and input to the Sin terminal 114. In the following, each of the combinations of the echo canceller convergence state and the magnitude of the echo path change due to transfer will be described.

(Problem 4: Extension transfer case A)
The extension transfer case A is when the convergence of the adaptive filter 118 is insufficient and the characteristics of the hybrid circuits 108 and 111 hardly change.

  In this case A before transfer, the echo cancellation residual e (n) in the equation (2) is not 0, so that the coefficient update in the equation (1) is executed while being valid. That is, as in the prior art, if x (n) is a tone signal, the coefficient of the adaptive filter 118 is updated according to the input x (n) according to the equation (1). As a result, the coefficient of the adaptive filter 118 converges so as to cancel the tone echo signal y1 rather than the characteristic of the hybrid circuit 108. Naturally, the coefficient of the adaptive filter 118 converges to a characteristic different from the impulse response of the hybrid circuit 108 as shown in FIG. Thereafter, transfer occurs, the hybrid circuit 111 is connected, and the telephone 110 is connected to start a call.

In this case, it can be considered that the hybrid circuit characteristics are the same before and after the transfer, but the coefficients of the adaptive filter 118 converge according to the tone. Thereafter, when the audio signal x (n), which is a non-tone signal, is input, the adaptive filter 118 estimates only the component of the tone signal. Therefore, as described in the item of the problem 2, the adaptive filter 118 The convergence state does not reflect the response characteristics of the hybrid circuit. Accordingly, the echo of an audio signal having a frequency component wider than the tone frequency cannot be canceled by almost all components and is output from the adder 119. The output of the adder 119 is input to a double talk detector (DTD) 117 as shown in FIG. The double talk detector 117 calculates the magnitude of the ratio ERLE between the adder output and the input rin (n) from the reception input terminal, and detects double talk based on the magnitude of the ERLE value or the magnitude of the change in ERLE. ERLE is calculated, for example, as shown in equation (3).

  As described in the prior art, if ERLE is large, such as 30 dB, e (n) is sufficiently small so that it can be converged or converged. If ERLE is small, such as 6 dB, e (n ) Is still large, the coefficient update of the adaptive filter 118 is stopped because there is a voice on the called side. Alternatively, as another method, a rapid deterioration of ERLE is detected, it is determined that there is a transmission voice on the called side, and the coefficient update of the adaptive filter is stopped.

  However, as described above, immediately after the tone misconvergence, since the echo component due to the audio signal from the reception input terminal is not canceled out by most components, ERLE has a small value, and also from the viewpoint of ERLE change, ERLE due to voice immediately after the tone deteriorates rapidly. The double talk detector 117 determines that there is a transmission voice signal on the called side, and detects that a double talk state has occurred. As a result, the double talk detector 117 stops updating the coefficient of the adaptive filter 118, so that the adaptive filter 118 freezes the coefficient thereafter and maintains it thereafter. That is, the coefficient update of the adaptive filter 118 is frozen in a state where the echo does not disappear. Further, if the hybrid circuits 108 and 111 have a response with a small initial delay, the coefficient of the adaptive filter 118 is not the same as the input tone characteristic. That is, the adaptive filter 118 reflects only characteristics that are different from the characteristics of the input signal and the characteristics of the hybrid circuits 108 and 111. As a result, it is not possible to recover from the misconvergence state.

  Fortunately, even if double-talk detection does not occur, the coefficients of the adaptive filter 118 converge to an inappropriate state for echo cancellation that reflects the tone signal rather than the characteristics of the hybrid circuits 108 and 111. Even if the characteristics of the hybrid circuit 111 are converged again, it is unsuitable as the initial setting for reconvergence of the adaptive filter 118. Therefore, it takes time to reconverge, and an echo is generated during that time. End up.

(Problem 4: Extension transfer case B)
The extension transfer case B is a case where the convergence of the adaptive filter 118 is insufficient and the characteristics of the hybrid circuits 108 and 111 are different.

  In case B, the problem is more noticeable than in case A. This is because after the tone misconvergence, the characteristics of the hybrid circuit 111 are completely different from the characteristics of the hybrid circuit 108, and the filter coefficients of the adaptive filter 118 once misconvergenced with the hybrid circuit 108 are newly added. Since the impulse response characteristic of the connected hybrid circuit 111 is not reflected at all, the echo of the voice having a wide frequency component after the tone signal cannot be canceled at all, and the echo becomes remarkable. Furthermore, the erroneous determination of the double talk detector 117 cannot be avoided. As a result, as described in Case A, the adaptive filter 118 freezes the coefficient update in a state where the echo does not disappear. Further, if the hybrid circuits 108 and 111 have a response with a small initial delay, the coefficient of the adaptive filter 118 does not become the same as the input tone characteristic, so that it cannot return from the frozen state.

  Accordingly, after the transfer, the state where the echo is generated is maintained.

(Problem 4: Extension transfer case C)
The extension transfer case C is a case where the convergence of the adaptive filter 118 is sufficient and the characteristics of the hybrid circuits 108 and 111 are hardly changed.

  In case C, since the echo removal residual e (n) in equation (2) is originally 0, the coefficient update according to equation (1) is not updated because the update amount is zero. Furthermore, if the step gain is reduced according to convergence as in the prior art, the coefficient update is prevented.

  However, in practice, the echo cancellation residual e (n) rarely becomes zero completely due to background noise on the near-end speaker (Sin speaker) side or the like. In such a case, when a tone signal is input from the reception input terminal 103 for a considerably long period of time, the coefficient of the adaptive filter 118 gradually converges to tone characteristics. At this time, once the adaptive filter 118 erroneously converges on the tone signal, the same problem as in case A occurs. In case C, it is desirable not to detect the result of misconvergence of the adaptive filter 118 as in the prior art, but to stop the coefficient update before the tone misconvergence of the adaptive filter 118 proceeds. .

(Problem 4: Extension transfer case D)
The extension transfer case D is when the adaptive filter 118 is sufficiently converged and the characteristics of the hybrid circuits 108 and 111 are different.

  Also in case D, when a tone signal for a long time is input from the telephone set 100, the coefficient of the adaptive filter 118 is erroneously converged by the tone signal. The problem that occurs immediately after such tone misconvergence is the same as in Case B. However, in case D, even if the coefficient update is stopped before the tone misconvergence of the adaptive filter 118 proceeds, the response characteristic of the hybrid circuit 111 after the extension transfer is the same as the characteristic of the hybrid circuit 108 before the transfer. Since they are different, anyway, echo cancellation is almost impossible after transfer. Therefore, ERLE has a small value. Even if it is regarded as a change, the way of ERLE change is also abruptly reduced with the occurrence of transfer, so the double talk detector 117 erroneously determines the change of ERLE as a double talk state. In addition, even if the coefficient of the adaptive filter 118 is stopped before the tone misconvergence progresses, the coefficient converged by the speech signal is frozen as it is. Even if the coefficients are analyzed, there is no tone property and ERLE is small, so it is not possible to escape from the double-talk misjudgment state. Therefore, the adaptive filter 118 of the echo canceller freezes the coefficient in a state where the echo cannot be erased, and continues to emit the echo after the transfer occurs. That is, even if misconvergence occurs, a problem such as Case B occurs, and even if coefficient updating is stopped before misconvergence, echoes continue to be output after transfer.

  The present invention has been made in consideration of the above points, and an object of the present invention is to provide an echo canceller that can reduce the feeling of echo in subsequent conversations even when a tone signal is input.

  The present invention relates to an echo canceller that uses an adaptive filter to generate a pseudo-echo and removes echo by a hybrid circuit. (1) Tone type determination for determining whether a far-end input signal is a predetermined type of tone signal And (2) a tone signal of a predetermined type determined from the far-end input signal to the adaptive filter when the tone type determination unit determines that the far-end input signal is a tone signal of a predetermined type. And a first band rejection filter provided at the input stage of the adaptive filter.

  Here, it is preferable to input the near-end input signal that has passed through the first band rejection filter to the double-talk detection means that detects the double-talk state of the far-end speaker and the near-end speaker.

  Also, there is an addition means for subtracting the pseudo echo signal from the adaptive filter from the near-end input signal, and the addition is performed when the tone type determination means determines that the far-end input signal is a predetermined type of tone signal. It is preferable to have a second band rejection filter provided at the input stage of the adding means for removing the determined frequency component of the tone signal of the predetermined type from the near-end input signal to the means.

  According to the present invention, it is possible to provide an echo canceller that can reduce the feeling of echo in subsequent conversations even when a tone signal is input.

It is a block diagram which shows the detailed structure of the echo canceller of 1st Embodiment. It is explanatory drawing (1) of the subject 2 of a prior art. It is explanatory drawing (2) of the subject 2 of a prior art. It is a block diagram for description of Problem 4 of the prior art. It is a block diagram which shows the structure of the telephone communication system including the echo canceller of 1st Embodiment. It is explanatory drawing of the determination operation | movement of the signal classification determination device of 1st Embodiment. It is a block diagram which shows the detailed structure of the echo canceller of 2nd Embodiment. It is explanatory drawing of the relationship between the determination result of the signal classification determination device of 2nd Embodiment, and an output. It is a block diagram which shows the detailed structure of the echo canceller of 3rd Embodiment. It is explanatory drawing of the relationship between the determination result of the signal type determination device of 3rd Embodiment, and an output. It is a block diagram which shows the detailed structure of the echo canceller of 4th Embodiment. It is a block diagram which shows the detailed structure of the echo canceller of 5th Embodiment.

(A) First Embodiment Hereinafter, a first embodiment in which an echo canceller according to the present invention is applied to an echo canceller provided in a VoIP terminal will be described with reference to the drawings.

  The echo canceller of the first embodiment has been made in view of the above-described problems 1 and 2 of the prior art, and is based on the fact that the actual harm caused by tone signal misconvergence occurs most often at the start of call setup. It was made. The echo canceller of the first embodiment is such that the initial value is appropriate when the echo canceller reconverges.

(A-1) Configuration of First Embodiment FIG. 5 is a block diagram showing a configuration of a telephone communication system including the echo canceller of the first embodiment.

  The telephone communication system shown in FIG. 5 is a system for executing communication by existing telephones 1 and 9 via the IP network 5, for example. The telephones 1 and 9 are accommodated in the corresponding VoIP terminals 4 and 6, and are connected to the IP network 5 via the VoIP terminals 4 and 6.

  In FIG. 5, the site A side is the called side and the site B side is the calling side, and FIG. 5 shows that the call signal that arrives from the site A to the site B is prevented from flowing from the site B to the site A as an echo. It is illustrated from the viewpoint of.

  The VoIP terminal 4 includes a hybrid circuit 2 connected to the telephone 1 with two lines, and a digital / analog converter (D / D) that converts a digital call signal directed to the telephone 1 to digital / analog and outputs the signal to the hybrid circuit 2. (A converter) 11 and an analog / digital converter (A / D converter) 3 for analog / digital conversion of the analog call signal output from the telephone set 1 via the hybrid circuit 2. Of course, an echo canceller may be mounted on the VoIP terminal 4.

  On the other hand, the VoIP terminal 6 includes the echo canceller 12 of the first embodiment, the hybrid circuit 8 connected to the telephone 9 with two wires, and a digital call signal output from the echo canceller 12 and directed to the telephone 9. A digital / analog converter (D / A converter) 7 that performs digital / analog conversion and outputs to the hybrid circuit 8 and an analog call signal output from the telephone 9 via the hybrid circuit 8 are analog / digital converted to an echo canceller 12. An analog / digital converter (A / D converter) 10.

  FIG. 1 is a block diagram showing the detailed configuration of the echo canceller 12 of the first embodiment provided at the position as described above together with the surrounding configuration. The same reference numerals are given to the same and corresponding parts as in FIG. As shown. In the echo canceller 12 of the first embodiment, the input / output signal is a digital signal.

  In FIG. 1, the echo canceller 12 of the first embodiment includes a reception input terminal Rin, a reception output terminal Rout, a transmission input terminal Sin, and a transmission output terminal Sout as input / output terminals. The echo canceller 12 includes an adder 13, a double talk detector (DTD) 14, a tone determination unit 15, a signal type determination unit 16, a coefficient controller 17, and an adaptive filter (ADF) 18.

  The adder 13 subtracts the pseudo echo signal y ′ from the signal from the transmission input terminal Sin to remove the echo component.

  The double talk detector 14 detects a talk state such as a double talk state based on the signal from the reception input terminal Rin and the output signal of the adder 13. The double talk detector 14 of the first embodiment also clears the internal state by a control signal from the coefficient controller 17.

  The tone determination unit 15 determines whether the signal input from the reception input terminal Rin is a signal having tone characteristics. The signal having tone characteristics refers to a repetitive waveform signal having a predetermined period such as various tones (for example, dial tone) for call control. The tone determination unit 15 may be of any type as long as it can determine whether or not it has tone characteristics. For example, the one described in JP 2000-295641 A is applied. obtain.

  The signal type determiner 16 determines whether the signal is a call control signal (tone signal related to call control) when the tone determiner 15 determines that the signal has tone characteristics. The determination method will be clarified in the description of the operation.

  The coefficient controller 17 determines whether or not to update the filter coefficient in the adaptive filter 18 according to the detection result of the double talk detector 14 and the determination result of the signal type determiner 16, and clears the filter coefficient. It is something to control.

  The adaptive filter 18 generates a pseudo echo signal y 'from the input signal from the reception input terminal Rin and an internal filter coefficient, and gives it to the adder 13 as a subtraction input. The adaptive filter 18 updates the filter coefficient while also taking into account the echo removal residual signal e. However, the adaptive filter 18 clears the filter coefficient when the coefficient controller 17 instructs to clear the filter coefficient.

(A-2) Operation of the First Embodiment Next, the operation of the echo canceller 12 of the first embodiment will be described together with the operation of the telephone communication system including the echo canceller 12.

  When the calling telephone 9 lifts the handset to call the speaker (not shown) of the telephone 1, a dial tone is output toward the telephone 9. The dial tone is output from, for example, an IP gateway device (not shown) between the VoIP terminal 6 and the IP network 5 in FIG. 5 and may be input to the reception input terminal Rin. There is also a case where a tone generator (not shown) provided is inputted to the reception input terminal Rin. In the following description, it is assumed that the signal is output from an IP gateway device (not shown) and then input to the reception input terminal Rin. The dial tone is a “two” sound that can be heard when a so-called handset is lifted, and is a tone signal having a frequency of 400 Hz in Japan.

  A reception signal (tone signal, speech signal, etc .; digital signal) input from the reception input terminal Rin is input to the double talk detector 14, the tone determination unit 15, the reception output terminal Rout and the adaptive filter 18.

  A reception signal (digital signal) input from the reception input terminal Rin and output as it is from the reception output terminal Rout is converted into an analog signal by the D / A converter 7, and the analog signal is converted to the telephone 9 via the hybrid circuit 8. Is output. However, a part of the converted analog signal is reflected by the hybrid circuit 8. The signal reflected by the hybrid circuit 8 is input to the transmission input terminal Sin via the A / D converter 10.

  The signal input to the transmission input terminal Sin is input to the double talk detector 14, the adaptive filter 18, and the transmission output terminal Sout after an echo component is appropriately removed via the adder 13. A signal is output from the transmission output terminal Sout toward the IP network 5 side.

  It is determined by the tone property determiner 15 whether the signal input to the tone property determiner 15 is a tone property signal. The tone characteristic determination unit 15 outputs a signal TON having a tone characteristic if the input signal is a tone characteristic signal, and outputs a signal TOFF having no tone characteristic to the signal type determination unit 16 otherwise. Is directly output to the signal type determination unit 16. If the signal input to the tone determination unit 15 is a dial tone, the tone determination unit 15 normally determines that there is a tone.

  In the signal type determination unit 16, for example, the frequency of the input signal is analyzed by a known FFT (Fast Fourier Transform), a known zero cross method, or the like. In the following description, it is assumed that the zero cross method, that is, the method of detecting the number of intersections of the signal amplitude with the horizontal axis is used.

  Here, assuming that the digital sampling frequency is 16 kHz, the input signal is a signal of 400 Hz when the number of samples T1 of the period (interval) that crosses zero in the same direction of the input signal waveform satisfies the equation (4). Is determined.

40−M1 ≦ T1 ≦ 40 + M1 (4)
40 in the equation (4) is 16 kHz / 400 Hz, and the period of zero crossing in the same direction (either negative to positive or positive to negative) when the input signal is a tone signal such as a sine wave signal. The number of samples. M1 is a parameter that defines an allowable error range. For example, 5 can be used. That is, when the number of samples T1 from the zero crossing to the next zero crossing satisfies the equation (4), it is determined that the input signal is a signal of 400 Hz. For signals of other frequencies f, the zero crossing interval is 16000 / f samples, and thus an equation for detecting the frequency may be determined in the same manner. The expression (5) is an expression corresponding to the expression (4) regarding the signal of the frequency f.

16000 / f−M1 ≦ T1 ≦ 16000 / f + M1 (5)
The signal type determiner 16 determines whether or not the input signal is a known call control signal from the tone presence / absence signals TON and TOFF given from the tone property determiner 15 and the frequency information obtained by frequency analysis. When the determination is made and the call control signal is determined, a coefficient reset signal RST is output to the coefficient controller 17 and the double talk detector 14. For example, the signal type determination unit 16 outputs a coefficient reset signal RST when a signal TON having tone characteristics is input and the frequency information obtained by frequency analysis is a predetermined frequency (for example, 400 Hz or 2100 Hz). For example, when a dial tone is input, since the predetermined frequency is 400 Hz, the signal type determination unit 16 outputs the coefficient reset signal RST even if the type of dial tone is not specified.

  FIG. 6 is an explanatory diagram showing the relationship between the call control signal (tone signal) and the frequency. The dial tone (DT), ringback tone (RBT), busy tone, and the like are tone signals of 400 Hz, and the FAX communication start tone is a tone signal of 2100 Hz. Therefore, for example, the signal type determination unit 16 stores 400 and 2100 in the internal memory, and when applying the equation (5), reads the stored 400 or 2100 as the frequency f, and (5) Make a decision according to the formula. FIG. 6 shows two types of predetermined frequencies, but three or more types of predetermined frequencies may be described.

  When the coefficient reset signal RST is input, the coefficient controller 17 outputs a signal CL_T that clears the coefficient of the adaptive filter 18 to the adaptive filter 18. The coefficient controller 17 does not output the clear instruction signal CL_T when the coefficient reset signal RST is not input. The adaptive filter 18 that has received the clear instruction signal CL_T clears the filter coefficient. The adaptive filter 18 re-executes the update of the filter coefficient from the initial state when the clear instruction signal CL_T does not arrive.

  The clear instruction signal CL_T signal is also given to the double talk detector 14 as described above. The double talk detector 14 detects double talk using ERLE, but when the clear instruction signal CL_T is given, the internal state is cleared.

  As described above, when the signal input to the reception input terminal Rin is a call control signal such as DT, the coefficient of the adaptive filter 18 is cleared. Actually, when the calling speaker on the telephone 9 side performs a dial operation and a human voice signal starts to flow through the call path, the signal type determination unit 16 does not output the coefficient reset signal RST. The filter 18 can start echo path estimation from the desired initial state of coefficient clear and can perform echo path estimation quickly. The double talk detector 14 calculates ERLE by using the cancellation addition output of the adaptive filter 18 that has started to converge from the cleared desired initial state and the input signal from the reception input terminal Rin, and performs correct double talk detection. Can be implemented.

(A-3) Effect of First Embodiment According to the first embodiment, a call control signal inevitably generated at the start of a call is detected, the coefficient of the adaptive filter is cleared, and double talk detection is performed. Since the internal state of the device is cleared, misconvergence due to tone signals (call control tones) can be prevented, and the update of the echo canceller's adaptive filter can be started from the desired initial state. It is possible to provide an echo canceller that is executed and has little performance degradation due to double talk. Here, even if the initial delay of the echo path is short, a good call without echo can be realized.

(B) Second Embodiment Next, a second embodiment in which the echo canceller according to the present invention is applied to an echo canceller provided in a VoIP terminal will be described with reference to the drawings.

  The echo canceller of the second embodiment has been made in view of the above-described problems 1, 2, and 3 of the prior art.

  Further, in the case where the first embodiment is applied, the present invention has been made in view of problems that occur when a tone detector (not shown in FIG. 1) is provided in the following position in the VoIP terminal 6. If the tone detector is not provided at the following position, the first embodiment functions effectively.

  When the echo canceller 12 is arranged in the VoIP terminal 6 as shown in FIG. 1 (first embodiment) described above, a tone (not shown) is provided on the output side of the transmission output terminal Sout regardless of the tone determination unit 15. A detector is often provided. This is because a push button signal (DTMF signal: PB signal) input from the telephone 9 is detected. However, when the output signal from the reception input terminal Rin is detected as a tone signal and the echo canceller is operated so as not to remove the echo as in the first embodiment, the following inconvenience occurs. To do.

  For example, a part of the dial tone signal input from the reception input terminal Rin is sent to the transmission input terminal Sin via the reception output terminal Rout, the D / A converter 7, the hybrid circuit 8, and the A / D converter 10. Input as echo y1. As described above, when the DTMF signal s1 that is a push button signal is output from the telephone set 9, the echo y1 and the DTMF signal s1 are added together by the hybrid circuit 8, and the echo canceller 12 passes through the A / D converter 10. Is output to the transmission output terminal Sout without echo cancellation. As a result, the above-described tone detector is disturbed by an extra echo signal y1 other than the DTMF signal, and inconveniences such as failure to detect the push button number occur. Such a problem occurs frequently in VoIP terminals that do not assume the above-mentioned inconvenience due to echo, and even when the dial button is pressed while lifting the handset and listening to the dial tone (two sounds), the telephone number is easily recognized. It causes a phenomenon that it cannot be made. In the worst case, it appears as a serious problem such as being unable to make a call.

  The second embodiment has been made in view of such a problem, and in addition to the same effects as those of the first embodiment, even if a signal such as a dial tone is input from the reception input terminal Rin, the telephone 9 Another object of the present invention is to provide an echo canceller that can detect a push button signal from the camera.

(B-1) Configuration of Second Embodiment FIG. 7 is a block diagram showing the detailed configuration of the echo canceller 12A of the second embodiment together with the surrounding configuration, and FIG. 1 according to the first embodiment described above. The same reference numerals are given to the same and corresponding parts.

  In FIG. 7, the echo canceller 12A of the second embodiment includes an adder 13, a double talk detector 14, a tone determination unit 15, a signal type determination unit 16A, and an adaptive filter 18 as in the first embodiment. Have. On the other hand, the echo canceller 12A of the second embodiment does not include the coefficient controller 17, but is provided with band rejection filters (NF) 21 and 22 instead, and the signal type determiner 16A also includes the first one. The processing is slightly different from that of the embodiment.

  The signal type determination unit 16A of the second embodiment determines not only the call control signal but also the specific type of the call control signal of the input signal. For example, the same 400 Hz dial tone and busy tone are distinguished and determined. For example, the signal type determination unit 16A of the second embodiment collates a spectrum pattern obtained by FFT or the like with a reference spectrum pattern, or collates a waveform pattern of an input signal with a reference waveform pattern after aligning the dynamic range. For example, the specific type of the call control signal is also determined. The signal type determination unit 16A gives type information to the band rejection filters 21 and 22 when the input signal is a tone signal and a specific type is obtained.

  The band rejection filter 21 is provided between the transmission input terminal Sin and the adder 16. On the other hand, the band rejection filter 22 is provided between the reception input terminal Rin and the double talk detector 14 and between the reception input terminal Rin and the adaptive filter 18.

  Each of the band rejection filters 21 and 22 is configured to prevent passage of a frequency component determined by the type information when the type information is given from the signal type determination unit 16A. That is, the band rejection filters 21 and 22 are band rejection filters that can vary the pass rejection band and the like.

(B-2) Operation of the Second Embodiment Next, the operation of the echo canceller 12A of the second embodiment will be described focusing on the differences from the first embodiment.

  When the signal type determination unit 16A of the second embodiment determines the specific type of the input signal, the signal type determination unit 16A outputs the determination result, that is, the signal KIND_TONE indicating the type of the signal, to both the band rejection filters 21 and 22. The signal type determiner 16A outputs nothing when the input signal is a non-tone signal.

  FIG. 8 is an explanatory diagram of the relationship between the determination result of the signal type determiner 16A and the output signal KIND_TONE. The signal type determiner 16A internally stores information representing the relationship as shown in FIG. 8, and outputs a signal KIND_TONE having a value corresponding to the determination result. In the example of FIG. 8, KIND_TONE = 1 represents a dial tone. The value of the signal KIND_TONE may conform to any system as long as it represents the frequency and the type of signal, and is different from that shown in FIG. 8, but for example, a numerical value 400 directly representing the frequency is directly represented by the signal KIND_TONE. The output value may be set as follows. Even when the input signal is a non-tone signal, a signal KIND_TONE having a value (for example, 0) indicating that the signal is a non-tone signal may be output.

  Each of the band rejection filters 21 and 22 to which the signal type determination signal KIND_TONE is input performs a known band rejection filter process on the input signal so as to block the frequency corresponding to the signal KIND_TONE. Each of the band rejection filters 21 and 22 allows the input signal to pass when there is no signal KIND_TONE.

  The band rejection filters 21 and 22 select one or more of a plurality of filters prepared in advance, that is, a band rejection filter that blocks a predetermined frequency (for example, 400 Hz) or other frequencies, and designates it with a signal KIND_TONE. A filter that blocks the input frequency may be realized, or a frequency blocking filter may be configured adaptively from time to time based on the input frequency. A method of adaptively realizing the band rejection filters 21 and 22 will be described. For example, if the rejection frequency indicated by the signal KIND_TONE is 400 Hz and the sampling frequency is 16 kHz, the input signal is 40 (= 16 kHz / 400 Hz). A filter that removes the 400 Hz tone can be realized by inverting the signal delayed by the sample (corresponding to a period of 400 Hz) and adding it to the input signal without delay. In short, any method may be used as long as a filter that removes a predetermined frequency can be realized in accordance with signal type and frequency information. As described above, the band rejection filters 21 and 22 prevent the call control signal from passing through the input signal type.

  The output signal from the band rejection filter 21 is input to the adder 13 and the echo is appropriately removed. On the other hand, the output signal from the band rejection filter 22 is input to the double talk detector 14 and the adaptive filter 18, and is used for detecting a double talk state and forming a pseudo echo signal y 'as appropriate.

  If the input signal from the reception input terminal Rin is determined to be a tone signal by the signal type determination unit 16A, the band rejection filter 22 removes the tone signal, so that the adaptive filter 18 and the double talk detector 14 include It is the same if nothing is entered. At this time, even if the DTMF signal s1 is output from the telephone set 9 toward the hybrid circuit 8, the signal after being band-stopped by the band-stop filter 21 is only the signal s1 and is input to the adder 13. On the other hand, the tone signal from the reception input terminal Rin is blocked by the band rejection filter 22 and nothing is input to the adaptive filter 18 and the double talk detector 14. That is, at this time, the input signal x (n) = 0 to the adaptive filter 18 is obtained, and the above-described equation (1) can be expressed as equation (6), and the coefficient update is substantially stopped.

h _k (n + 1) = h _k (n + 1) +0 (6)
In addition to such coefficient update control, the sum of the squares of the input signals as denominators ≦ MIN (dBm0) so that the denominator of the second term on the right side of equation (1) = 0 and division does not diverge. ) × multiply-sum number, updating is stopped (0 dBm0 reference value of dBm0 = 0 dBm0 is as described in the well-known ITU (International Telecommunication Union) -T (communication sector) standard recommendation G.711). For example, MIN = −50 dBm0 can be applied, but is not limited to this and may be set as appropriate.

  When the band rejection filters 21 and 22 are used as described above, the band rejection filters 21 and 22 block the tone signal from the reception input terminal Rin and the echo due to the tone. Even if it is executed for a time, the actual update amount becomes 0, and the coefficient value itself is not updated and does not change. Further, since the adaptive filter 18 whose reference input is 0 outputs nothing, no extra pseudo echo y ′ is output to the adder 13. Therefore, the DTMF signal s1 input to the adder 13 is not subjected to any deterioration, and only the echo y1 is removed and output to the transmission output terminal Sout. An output signal (DTMF signal) from the transmission output terminal Sout is correctly detected by a tone detector (not shown).

  In addition, since there is no false convergence in the tone signal, the internal state of the double talk detector 14 is not disturbed, and the double talk detection in the normal voice following the tone can be performed with high accuracy and the echo can be removed.

(B-3) Effects of Second Embodiment As described above, according to the second embodiment, the following effects can be obtained in addition to the same effects as those of the first embodiment. Even if there is a tone signal from the receiving input terminal Rin, and the accommodating telephone outputs a tone signal such as a DTMF signal in the opposite direction, the tone detector using the echo canceller output correctly identifies the DTMF signal and makes a call. Because there is no misconvergence in the tone signal regardless of the initial delay of the echo path, the internal state of the double talk detector is not disturbed, and the double talk detection in the normal voice following the tone is also accurate. You can go well and remove the echo.

(C) Third Embodiment Next, a third embodiment in which the echo canceller according to the present invention is applied to an echo canceller provided in a private branch exchange will be described with reference to the drawings.

  The echo canceller of the third embodiment, when applied to a large-scale system with echo path switching such as PBX connection in which extension exchange occurs, automatically escapes from the state where echo cancellation is not possible and has no echo. This is intended to achieve a telephone call, and has been made in view of the above-described problems 1 to 4 of the prior art.

(C-1) Configuration of Third Embodiment FIG. 9 is a block diagram showing the detailed configuration of the echo canceller of the third embodiment together with the surrounding configuration, and the first and second embodiments already described. The same and corresponding parts as those in FIGS. 1 and 7 are indicated by the same reference numerals.

  In FIG. 9, the echo canceller 12B of the third embodiment is provided in a private branch exchange (PBX) 6B having an extension transfer function. The private branch exchange 6B, for example, accommodates the telephones 9-1 and 9-2 before and after the transfer, and goes to the hybrid circuits 8-1 and 8-2 that perform 4-wire 2-line conversion, and the telephones 9-1 or 9-2. Digital / analog converter (D / A converter) 7 for digital / analog conversion of digital signal, analog / digital converter (A / D conversion) for analog / digital conversion of signal output from telephone set 9-1 or 9-2 10), the switch 30-1 for supplying the output signal of the D / A converter 7 to the hybrid circuit 8-1 or 8-2, and the signal output from the hybrid circuit 8-1 or 8-2 as A / D The switch 30-2 given to the converter 10 and the echo canceller 12B of the third embodiment are provided. The switches 30-1 and 30-2 are switched in conjunction with each other.

  The echo canceller 12B according to the third embodiment includes an adder 13, a double talk detector 14B, a tone determination unit 15B, a signal type determination unit 16B, a coefficient controller 17B, an adaptive filter 18, band elimination filters 21, 22 and a filter. A state determination unit 31 and an echo cancellation amount calculator (ACANC) 32 are included.

  The adder 13, the adaptive filter 18, and the band rejection filters 21 and 22 are the same as those in the second embodiment.

  The tone property determination unit 15B of the third embodiment can also determine two types of synthesized tone signals such as a DTMF signal in addition to the functions of the above-described embodiments.

  The signal type determiner 16B of the third embodiment can also determine the types of two types of synthesized tone signals such as DTMF signals, using the determination result of the tone property determiner 15B. The signal type determiner 16B also determines the end of the tone signal.

  The echo cancellation amount calculator 32 calculates the echo cancellation amount from the input / output signal of the adder 13, in other words, from the input signal having an echo component and the output signal after the echo removal operation is performed. .

  The filter state determiner 31 determines the state of the adaptive filter 18 from the determination result of the signal type determiner 16B and the calculation result of the echo cancellation amount calculator 32.

  The coefficient controller 17B of the third embodiment determines control contents such as stop of coefficient update of the adaptive filter 18 and clear of the coefficient from the determination result by the filter state determiner 31.

  The double talk detector 14B of the third embodiment controls the coefficient update for the adaptive filter 18 according to the control content of the coefficient controller 17B and the detection result of the double talk in itself.

(C-2) Operation of the Third Embodiment Next, the operation of the echo canceller 12B of the third embodiment will be described. In the following, a description will be given of a situation in which a transfer for switching from the called telephone 9-1 to another telephone 9-2 on the called side occurs. Here, since the description is before and after the occurrence of the transfer, it is assumed that the calling telephone and the called telephone 9-1 (not shown) have already made a call.

  First, a voice signal output from a calling telephone (not shown) is converted into a digital signal by an A / D converter (not shown) in the previous stage of the echo canceller 12B and input to the reception input terminal Rin. The signal input to the reception input terminal Rin is provided to the band rejection filter 22 and the tone determination unit 15B, and is output as it is from the reception output terminal Rout and input to the D / A converter 7.

  The signal converted into an analog signal by the D / A converter 7 is input to the hybrid circuit 8-1 through the switch 30-1, and is given to the telephone set 9-1 through the hybrid circuit 8-1. , A part of the signal is reflected by the hybrid circuit 8-1 and becomes an echo signal y1. The echo signal y1 is given to the A / D converter 10 via the switch 30-2, converted into a digital signal again, and inputted to the transmission input terminal Sin of the echo canceller 12B.

  A signal input to the transmission input terminal Sin is input to the band rejection filter 21. As will be described later, each of the band rejection filters 21 and 22 removes the tone signal according to the output of the signal type determiner 16B, outputs the signal Sin_AC after tone removal to the echo cancellation amount calculator 32, and adds Output to the device 13. The adder 13 adds the signal Sin_AC after the tone removal and the pseudo echo y ′ to remove the echo. The output signal res from the adder 13 is input to the double talk detector 14B and the echo cancellation amount calculator 32, and is transmitted from the transmission output terminal Sout to a remote calling side telephone (not shown).

Here, the operation of the echo cancellation amount calculator 32 will be described. The echo cancellation amount calculator 32 calculates the echo cancellation amount ACANC (n), for example, according to equation (4). In the formula (4), n represents the nth calculated value.

In the equation (4), the square ratio of Sin_AC (n) and res (n) is directly logarithmically calculated. However, if the emphasis is placed on the rough behavior of the signal, the change in the characteristic may be monitored. , (5), Sin_AC and res may be collected for each predetermined number of samples (for example, 160 samples), and the sum of the collected samples may be used to calculate the logarithm of both. Further, an absolute value may be used instead of the square. In the equation (5), M is the number of samples to be averaged, and may be 160, for example, but is not limited to this number of samples.

  In short, as long as the level ratio and power ratio of the signals before and after the adder 13 are calculated, any calculation method by the echo cancellation amount calculator 32 may be used. The echo cancellation amount ACANC (n) obtained by the equation (4) is obtained by calculating the square of the signals before and after the adder 13 by a logarithmic ratio, and the output signal res from the adder 13 according to the convergence of the echo canceller 12B. As becomes smaller, the echo cancellation amount ACANC (n) becomes larger. As an initial value of the echo cancellation amount ACANC (n), 0 can be applied, but is not limited to this. The echo cancellation amount calculator 32 inputs the calculated echo cancellation amount ACANC (n) to the filter state determination unit 31.

  In the third embodiment, the echo cancellation amount is calculated by the echo cancellation amount calculator 32 using the signals before and after the adder 13 in order to calculate the echo removal amount. An example is as follows. A signal may be taken from an arbitrary point between the reception input terminal Rin and the reception output terminal Rout and a point immediately after the adder 13, and the echo attenuation amount including the attenuation amount of the echo path may be calculated.

  The filter state determination unit 31 is provided with an echo cancellation amount ACANC (n) from the echo cancellation amount calculator 32, and a band rejection filter control signal KIND_TONE and a tone end signal TEND described later from the signal type determination unit 16B. .

  Hereinafter, operations of the tone determination unit 15B and the signal type determination unit 16B according to the third embodiment will be described.

  The output signal of the tone property determiner 15B is input to the signal type determiner 16B. The tone determination unit 15B according to the third embodiment can determine a DTMF signal, that is, two types of synthesized tone signals, in addition to the function of the tone determination unit according to the second embodiment. The tone property determination unit 15B is configured to divide a signal into two broad bands in advance so that two types of synthesized tone signals can be determined. This is based on the fact that the DTMF signal is generated by synthesizing two kinds of tones of “high group” composed of high frequencies and “low group” composed of low frequencies.

  In the third embodiment, the tone determination unit 15B separates the “low group” from the bandpass filter having a pass band from 1000 Hz to 1700 Hz in order to separate the “high group”. A band pass filter having a pass band from 600 Hz to 980 Hz and a pass band from 0 Hz to 500 Hz for separating the “call control signal group” in the vicinity of 400 Hz as described in the first and second embodiments. The tone characteristic determination unit 15B performs the same tone frequency determination by zero crossing as described in the first embodiment for the signals after band division by these band pass filters. Has been made to do.

  The signal type determination unit 16B of the third embodiment determines the signal type for signals in the frequency band of the “call control signal group” as in the second embodiment described above (see FIG. 8). For a signal having both the “group” and “low group” frequency bands, the type of the signal is determined based on the relation information between the input, the determination result, and the output as shown in FIG. In addition, when "call control signal group" and "high group" are detected, when "call control signal group" and "low group" are detected, "call control signal group" and "high group" ”And“ low group ”may be handled as not being a tone signal, but the present invention is not limited to this.

  When the signal type determiner 16B determines the signal type, it outputs the determination result, that is, the signal KIND_TONE indicating the signal type to the band rejection filters 21 and 22 and the filter state determiner 31. The signal KIND_TONE can be any signal as long as the frequency and the signal type are associated, for example, the value (number) identifying the signal and the signal type are associated as shown in FIGS. It may have a value system.

  In the example of FIGS. 8 and 10, when KIND_TONE = 1, it represents a dial tone (DT), and when KIND_TONE = 5, it represents “1” of the DTMF signal.

  Each of the band rejection filters 21 and 22 to which the signal KIND_TONE is input from the signal type determination unit 16B performs a known band rejection filter process on the input signal so as to block a frequency determined according to the signal KIND_TONE. The behavior of the band rejection filters 21 and 22 is the same as that of the second embodiment, but in the case of this third embodiment, when the signal KIND_TONE indicates the DFMF signal, two types of signals are removed. The point is different from that of the second embodiment. For example, a filter for “high group” and “low group” determined according to the signal KIND_TONE is selected from a plurality of band rejection filters prepared in advance for “high group” and “low group”, respectively. In this way, the band removal according to the DTMF signal (which is cascaded) is performed. The configuration method for blocking the band of the DTMF signal is not limited to this method.

  In the case of the third embodiment, the signal type determiner 16B also outputs the signal KIND_TONE to the filter state determiner 31. Further, the signal type determiner 16B outputs a tone end signal TEND to the filter state determiner 31 when the signal from the tone property determiner 15B changes from the tone property signal TON to the tone property absent signal TOFF.

  The filter state determiner 31 operates as follows according to the output signals KIND_TONE and TEND from the signal type determiner 16B and the output signal from the echo cancellation amount calculator 32.

  When the signal KIND_TONE is not input from the signal type determiner 16B, that is, during the period when the input signal from the reception input terminal Rin has no tone property, the filter state determiner 31 receives the signal from the echo cancellation amount calculator 32. The output signal is held while being updated at predetermined time intervals. For example, update holding is executed every 20 ms, but the update interval is not limited to this.

  On the other hand, when the signal KIND_TONE is input from the signal type determiner 16B, the filter state determiner 31 operates by classifying the output signal of the echo cancellation amount calculator 32 into conditions 1 to 4 as follows. .

Condition 1: ACANC (n) ≧ E_ACANC dB
Condition 2: ACANC (n) <E_ACANC dB
Note that, for example, 20 dB is applied as the threshold parameter E_ACANC that defines Condition 1 and Condition 2. However, the value of the threshold parameter E_ACANC is not limited to this value.

  When the condition 1 is satisfied, in other words, when the echo cancellation amount is large, the filter state determiner 31 outputs a signal ADP_STP that prompts the coefficient controller 17B to stop updating the coefficient of the adaptive filter 18. The coefficient controller 17B outputs the signal ADP_STP to the adaptive filter 18 via the double talk detector 14B, and stops the coefficient update of the adaptive filter 18.

  When the period of the tone signal ends and the signal TEND from the signal type determiner 16B is input to the filter state determiner 31, the filter state determiner 31 once stops updating the echo cancellation amount ACANC (n). Wait until the next echo cancellation amount ACANC update period comes. Then, when the echo cancellation amount ACANC (n + 1) is newly calculated by the echo cancellation amount calculator 32 and output to the filter state determination unit 31, the filter state determination unit 31 obtains the echo cancellation amount ACANC (n + 1). Comparison with previously held ACANC (n) is performed.

  That is, the echo cancellation amount ACANC (.) Before and after the tone signal is input to the echo canceller 12B is compared. (.) Represents an arbitrary n. Thereafter, following condition 1, determinations of conditions 3 and 4 described later are executed, and an operation corresponding to each result is executed.

  On the other hand, when the condition 2 is satisfied, in other words, when the echo cancellation amount is small, the filter state determination unit 31 does not output the signal ADP_STP that prompts the coefficient controller 17B to stop the coefficient update of the adaptive filter 18.

  The filter state determiner 31 determines the following conditions 3 and 4 and operates according to the result. In addition, although 3 dB can be applied as Δ1 and 0 dB as Δ2 in the conditions 3 and 4, it is not limited to these.

Condition 3: Δ2 <ACANC (n + 1) <ACANC (n) −Δ1
Condition 4: ACANC (n + 1) <Δ2
When the condition 3 is satisfied, in other words, a tone signal is input, but when the echo cancellation amount is much smaller than before input, the filter state determination unit 31 outputs the coefficient update promotion signal ADP_F to the coefficient controller 17B. To do.

  When the condition 4 is satisfied, in other words, when the echo cancellation amount becomes negative due to the echo removal operation, the coefficient reset signal ADP_RST is output to the coefficient controller 17B.

  When neither condition 3 nor condition 4 is satisfied, the filter state determination unit 31 does not output anything to the coefficient controller 17B.

  Instead of the above condition 3, a modification of the following condition 3 may be applied. In the modified example of condition 3, a margin width is set up and down using ACANC (n) instead of the lower limit fixed value Δ2.

Modification of condition 3:
ACANC (n) −Δ3 <ACANC (n + 1) <ACANC (n) −Δ1
When the coefficient reset signal ADP_RST is input from the filter state determination unit 31, the coefficient controller 17B outputs a reset signal RST to the double talk detector 14B. The reset signal RST is given to the adaptive filter 18 via the double talk detector 14B, and the adaptive filter 18 resets the filter coefficient and then restarts the coefficient update. Here, both the double talk detector 14B and the adaptive filter 18 are reset, but only the adaptive filter 18 may be reset.

  The coefficient controller 17B resets the double talk detector 14B when the coefficient update promotion signal ADP_F is input from the filter state determiner 31, but does not reset the adaptive filter 18 and forces the coefficient update operation. Control to run.

  In the third embodiment, the coefficient controller 17B controls the adaptive filter 18 via the double talk detector 14B. However, the coefficient controller 17B controls the adaptive filter 18 and the double talk detector 14B. Each may be controlled.

  How the operation of the third embodiment solves the problem 4 in the prior art will be described. Here, as described in the section of the problem, the description will be made according to the combination of the quality of the convergence state of the adaptive filter of the echo canceller at the time of transfer and the change in the echo path characteristics before and after the transfer.

  The most typical example in which extension transfer occurs is that a far-end speaker (not shown) (calling speaker in FIG. 9) performs a transfer operation on the push button (DTMF signal is output), Thereafter, the transfer function of the private branch exchange (PBX) 6B changes the telephone on the called side from the telephone set 9-1 to the telephone set 9-2. In other words, this is a case where a push button for designating a telephone 9-2 different from the telephone 9-1 connected first is operated and actually reconnected to another telephone 9-2.

(Extension transfer case A)
The extension transfer case A is a case in which the convergence of the adaptive filter 18 is insufficient and the response characteristics of the hybrid circuits 8-1 and 8-2 before and after the transfer hardly change.

  Some calls are made before transfer, and the adaptive filter 18 is in the middle of convergence due to the voice. Or, if some coefficient disturbance is received before the tone detection, this case A is applicable.

  When the convergence of the adaptive filter 18 is insufficient, the far-end speaker requests connection to a different telephone set 9-2, for example, by operating a push button (inputting a DTMF signal).

  The DTMF signal input to the reception input terminal Rin is input to the band rejection filter 22 and the tone determination unit 15B, and is output as it is from the reception output terminal Rout and is supplied to the D / A converter 7.

  The tone property determiner 15B detects the tone property. A signal KIND_TONE is output from the signal type determiner 16B in response to the output of the tone property determiner 15B, and the band rejection filters 21 and 22 block the passage of a frequency corresponding to the signal KIND_TONE.

  As a result, since the DTMF signal is blocked by the band rejection filter 21 while the DTMF signal is generated, the adaptive filter 18 and the double talk detector 14B are in the same state as when no DTMF signal is input. ing. Therefore, even if the DTMF signal is input to the reception input terminal Rin for a long time, the coefficient of the adaptive filter 18 is not disturbed, but the convergence of the coefficient does not proceed to the optimum state.

  Case A is when the adaptive filter 18 is not sufficiently converged, and the echo cancellation amount is small, so condition 2 is satisfied (when condition 1 is satisfied, it will be described later). When the DTMF signal ends, a tone end signal TEND is output from the signal type determiner 16B. Upon receiving this signal TEND, the filter state determination unit 31 compares the echo consumption amount ACANC (n) previously held with the echo cancellation amount ACANC (n + 1) newly output from the echo cancellation amount calculator 32. In this case A, the characteristics of the hybrid circuit 8-1 and the characteristics of the hybrid circuit 8-2 are almost the same. Therefore, when the adaptive filter 18 executes coefficient update, ACANC (n + 1) ≧ ACANC (n). . Therefore, the coefficient controller 17B outputs nothing.

  When the condition 1 is satisfied, the filter state determination unit 31 once stops the coefficient update of the adaptive filter 18 via the coefficient controller 17B, but the fact that the condition 1 is satisfied in the first place is that before the DTMF signal. This means that the characteristics of the hybrid circuit could be estimated from the signal. Since the case A is when the characteristics of the hybrid circuit 8-1 and the hybrid circuit 8-2 before and after the transfer are equal, the adaptive filter 18 desirably proceeds to update the coefficient as it is, and the coefficient controller 17B Is not output. As a result, the adaptive filter 18 and the double talk detector 14B can proceed with the normal echo removal operation as desired, and can proceed with the echo removal without deterioration of the echo before and after the generation of the tone signal.

(Extension transfer case B)
The extension transfer case B is a case in which the convergence of the adaptive filter 18 is insufficient and the characteristics of the hybrid circuits 8-1 and 8-2 change before and after the transfer.

  Also in case B, the DTMF signal ends, the tone end signal TEND is output from the signal type determiner 16B, and the filter state determiner 31 that has received this signal TEND newly adds the previously held echo cancellation amount ACANC (n). The operation of each unit until the echo cancellation amount ACANC (n + 1) output from the echo cancellation amount calculator 32 is compared is the same as in case A.

  When the signal KIND_TONE is input from the signal type determiner 16B, the filter state determiner 31 determines whether the condition 1 or the condition 2 is satisfied based on the output from the echo cancellation amount calculator 32. Since the convergence of the adaptive filter 18 is insufficient, in this case B, in most cases, the condition 2 is satisfied (when the condition 1 is satisfied, it will be described later).

  In addition, in this case B, the response characteristic of the hybrid circuit changes from the characteristic of the hybrid circuit 8-1 that does not match to the characteristic of the hybrid circuit 8-2 at the transfer. Even if the adaptive filter 18 tries to update the coefficient, the target of the echo path estimation has changed from the hybrid circuit 8-1 to the hybrid circuit 8-2. Naturally, the pseudo echo generated in the adaptive filter 18 becomes inappropriate, and ACANC (n + 1) <ACANC (n). In many cases, ACANC (n + 1) <0.

  Therefore, regardless of whether the determination result of the filter state determination unit 31 is the condition 1 or the condition 2, the subsequent condition determination is always either the condition 3 or 4. Further, the double talk detector 14B erroneously determines the echo after the path change as a near-end speaker signal due to a rapid deterioration in echo consumption, determines the state as double talk, and determines the coefficient of the adaptive filter 18. Stop updating.

  As described above, the filter state determiner 31 determines whether the change in the echo cancellation amount ACANC before and after receiving the output TEND from the signal type determiner 16B matches the condition 3 or 4, and first, double-talk The detector 14B is reset to cancel the erroneous determination of the double talk detector 14B. Next, the filter state determination unit 31 controls the adaptive filter 18, but changes the control according to the conditions 3 and 4 as described below.

a) When Condition 3 is satisfied The fact that Condition 3 is satisfied indicates that the echo cancellation amount ACANC has deteriorated within a predetermined allowable range. Therefore, although it is possible to remove some echoes, it is desirable to further perform convergence for optimal echo path estimation. Therefore, the filter state determination unit 31 outputs the coefficient update promotion signal ADP_F to the coefficient controller 17B to update the filter coefficient of the adaptive filter 18.

b) When the condition 4 is satisfied When the parameter Δ2 in the condition 4 is 0 dB, for example, the condition 4 is satisfied. This means that the echo cancellation amount ACANC is negative, that is, the echo cancellation fails, rather, the echo is emphasized. It shows that you are doing. In this case, for optimal echo path estimation, it is desirable to temporarily discard the coefficients of the adaptive filter 18 and execute the reconvergence operation. Accordingly, the coefficient controller 17B outputs the coefficient reset signal ADP_RST to the adaptive filter 18 via the double talk detector 14B, resets the filter coefficient, and restarts the coefficient update. In the third embodiment, the adaptive filter 18 is reset via the double talk detector 14B. However, the coefficient controller 17B may reset the adaptive filter 18 directly.

  Since the adaptive filter 18 is controlled in accordance with the above conditions 3 and 4, even if the characteristics of the hybrid circuit change before and after the transfer, the echo canceller 12B can quickly determine the characteristics of the newly connected hybrid circuit 8-2. Because it follows again, it is possible to make a call with the echo quickly removed.

(Extension transfer case C)
The extension transfer case C is a case where the convergence of the adaptive filter 18 is sufficient and the characteristics of the hybrid circuit hardly change before and after the transfer.

  In this case C, since the convergence of the adaptive filter 18 is sufficient, the condition 1 is satisfied in the filter state determination unit 31. The estimated echo path can uniformly estimate the voice frequency band, and naturally the echo of the tone signal as well as the voice signal can be removed. On the other hand, since the echo cancellation residual res is also almost 0, the coefficient update amount in the equation (1) is also 0, and even if a tone signal is inputted, the coefficient update is almost stopped.

  Exceptionally, if the DTMF signal is input for a long time (for example, 1 minute), the coefficient of the adaptive filter 18 may be gradually destroyed. In case C, this may be prevented.

  As described above, the combination of the filter state determination unit 31, the signal type determination unit 16B, and the band rejection filters 21 and 22 plays this role. When the condition 1 is satisfied with the signal KIND_TONE being output from the signal type determiner 16B, the filter state determiner 31 outputs a signal ADP_STP that prompts the coefficient controller 17B to stop updating the coefficient of the adaptive filter 18. Therefore, the coefficient update of the adaptive filter 18 is stopped, and the coefficient update is resumed after the transfer, so that the echo can be continuously removed without any problem.

  In the transfer case C, the same operation according to the conditions 3 and 4 is performed thereafter. However, in the case C, it can be considered that the characteristics of the hybrid circuit before and after the transfer do not change. Therefore, in terms of response characteristics, the hybrid circuit 8-1 and the hybrid circuit 8-2 are substantially equal, so that the coefficient of the adaptive filter 18 is not disturbed unless the input of the DTMF signal is so long. There is no deterioration in the echo cancellation amount ACANC. Therefore, there is no problem even if the echo canceller 12B continues to operate as it is.

  Even if the DTMF signal is input for a long time, the filter state determination unit 31 that has received the output KIND_TONE of the signal type determination unit 16B prevents the coefficient disturbance of the adaptive filter 18 thereafter. If the tone type determination is delayed, and the coefficient update stop of the adaptive filter 18 by the filter state determination unit 31 by the input of the signal KIND_TONE is delayed, the coefficient of the adaptive filter 18 is disturbed. Since the echo cancellation amount ACANC deteriorates at the boundary, the filter state determination unit 31 to which the signal TEND is input from the signal type determination unit 16B after the end of the DTMF signal determines again with the conditions 3 and 4 of the echo cancellation amount ACANC. Thereafter, the operation is the same as in cases A and B. That is, at worst, after the transfer, the adaptive filter 18 can be quickly reconverged and the echo can be removed as in the cases A and B described above.

(Extension transfer case D)
The extension transfer case D is a case where the convergence of the adaptive filter 18 is sufficient, and the characteristics of the hybrid circuit change at the transfer.

  In this case, the condition 1 is satisfied since the echo canceller 12B has sufficiently converged before the transfer.

  Therefore, as described above, the coefficient controller 17B temporarily stops the coefficient update of the adaptive filter 18 according to the output of the filter state determiner 31, but after the passage of the DTMF signal, Condition 3 or Condition 4 is satisfied. The convergence of the adaptive filter 18 is resumed, and after the transfer, the adaptive filter 18 can start reconvergence immediately and remove the echo.

(C-3) Effect of Third Embodiment As described in detail above, according to the third embodiment, there is a tone signal such as a dial tone from the reception input terminal, and the telephone set accommodated in the opposite direction. Even if a DTMF signal is output to the tone detector, the tone detector using the output of the echo canceller can correctly identify the DTMF signal, determine the telephone number, and make a call. Even if the echo path is re-estimated quickly, there is little performance degradation due to double talk, and the initial delay of the echo path is small, it can be reconverged quickly and a good call without echo can be realized. In addition to these effects, even when applied to a large-scale system where extension transmission occurs, the echo cancellation performance is not degraded by extension transfer, so excellent voice quality is achieved. An echo canceller that can be provided can be realized.

(D) Fourth Embodiment Next, a fourth embodiment in which the echo canceller according to the present invention is applied to an echo canceller provided in a private branch exchange will be described with reference to the drawings.

  The fourth embodiment has been made in view of the fact that in human-to-human calls, human voices may occasionally exhibit tone characteristics.

(D-1) Configuration of the Fourth Embodiment FIG. 11 is a block diagram showing the detailed configuration of the echo canceller of the fourth embodiment together with the surrounding configuration, and FIG. 9 is related to FIG. 9 according to the third embodiment. The same and corresponding parts are denoted by the same reference numerals.

  In FIG. 11, the echo canceller 12C of the fourth embodiment has a third feature in that the component parts of the tone property determination unit 15B and the signal type determination unit 16B of the third embodiment are replaced with a signal component determination unit 40. This embodiment is different from the third embodiment, and the rest is the same as the third embodiment.

  The signal component determination unit 40 forms the signals KIND_TONE and TEND based on the input signal from the reception input terminal Rin. The formation method will be clarified in the description of the operation.

(D-2) Operation of the fourth embodiment The fourth embodiment is different from the third embodiment only in the operation of the signal component determiner 40. The description about is omitted.

  The signal component determination unit 40 determines whether the input signal is a wideband signal (the input signal component is wideband) regardless of whether the input signal is a known call control signal (for example, DT, DTMF, etc.) or not. Or not). The signal component determiner 40 operates to determine, for example, a tone signal such as 1500 Hz that is not a call control signal as a tone signal.

  Hereinafter, a determination method in the signal component determination unit 40 will be described.

  The signal component determination unit 40 performs, for example, fast discrete Fourier transform (FFT) on the input signal and decomposes it into individual single frequency components. For example, an input signal is converted into a frequency component using 256-point FFT. If the sampling frequency is 16 kHz and 256-point FFT is used, each frequency component can be calculated with 128 resolutions for the 0 to 8 kHz band of the input signal, and the power spectrum P_f (k) can be obtained. In the fourth embodiment, the power P_min_f of the smallest frequency min_f in each power spectrum is regarded as the noise floor frequency power level, and when the following condition 5 is satisfied, it is regarded as “there is a frequency component”.

Condition 5: P_f (k)> P_min_f + δf
Here, δf is a determination offset value, and 15 dB can be applied, but is not limited to this value. k represents the number of the decomposed frequencies, and can take 128 values of 0 ≦ k ≦ 127.

  The signal component determination unit 40 counts the number of frequencies f (k) satisfying the condition 5 with a built-in counter, and when the count result C_F is C_F <TH_VOICE, the input signal is regarded as an arbitrary tone signal. The signal KIND_TONE is output to the filter state determiner 31. For example, 4 can be applied as the threshold parameter TH_VOICE, but is not limited thereto.

  In the fourth embodiment, since it is not necessary to determine the type of the tone signal, information for forming the signal KIND_TONE is stored in advance as in the first to third embodiments, and the information is referred to. The only problem is whether or not the signal KIND_TONE is output.

  In addition, when it is necessary to assign a number to the signal KIND_TONE in relation to the configurations of the band rejection filters 21 and 22 and the filter state determination unit 31, for example, numbers that are not applied in FIG. 8 and FIG. 10 described above are used. .

  In the case of the fourth embodiment, since the frequency of the detected tone signal is not determined in advance, the band rejection filters 21 and 22 may be subjected to a filter operation by applying a signal KIND_TONE incorporating frequency information.

  When the signal component determination unit 40 changes from the state in which the tone signal is detected according to the condition 5 as described above to the state in which the tone signal cannot be detected, the signal component determination unit 40 sends the signal TEND indicating the end of the tone signal to the filter state determination unit 31. Output.

(D-3) Effect of the Fourth Embodiment According to the fourth embodiment, an error occurs with respect to an arbitrary tone signal that is not classified as a call control signal (for example, an accidental human tone-like voice in a call between humans). Convergence and subsequent malfunction of the echo canceller can be prevented.

  An arbitrary tone signal is input when, for example, music is applied at the same level as the speaker's voice as the background sound of one or both speakers, or when the speaker sings during a call. It occurs very rarely when the individual person's voice frequency characteristics have specific tone characteristics. The generated tone frequency is not limited to the frequency of the call control tone, and is arbitrary, so a known frequency table cannot be referred to. In the fourth embodiment, a signal component determination unit 40 is provided to detect the tone characteristics of the input signal even if the frequency is arbitrary. After the detection result, as in the third embodiment, the adaptive filter and Since the double talk detector is controlled, the adaptive filter is prevented from misconvergence, and even if the initial delay of the echo path is small, it returns quickly from the misconvergence state. As a result, it is possible to realize an echo canceller capable of continuing to remove the echo without deteriorating.

(E) Fifth Embodiment Next, a fifth embodiment in which the echo canceller according to the present invention is applied to an echo canceller provided in a private branch exchange will be described with reference to the drawings.

  The echo canceller of the fifth embodiment is intended to make the hardware scale smaller than those of the third and fourth embodiments.

(E-1) Configuration of Fifth Embodiment FIG. 12 is a block diagram showing the detailed configuration of the echo canceller of the fifth embodiment together with the surrounding configuration, and is a block diagram showing FIG. 9 according to the third embodiment. The same and corresponding parts are denoted by the same reference numerals.

  In FIG. 12, the echo canceller 12D of the fifth embodiment is different from the third embodiment in that the band rejection filters 21 and 22 are not provided, and the function of the signal type determination unit 16D is the third embodiment. It is different that it is slightly different from the ones.

(E-2) Operation of Fifth Embodiment Next, an operation of the echo canceller 12D of the fifth embodiment that is different from that of the third embodiment will be described.

  A situation when a transfer occurs between the receiving side telephones 9-1 and 9-2 will be described. Since the description is before and after the transfer is generated, the calling side telephone and the called side telephone 9-1 (not shown) will be described as being already in a call state.

  The signal type determiner 16D outputs a signal TEND indicating the end of the tone signal to the filter state determiner 31 when the output of the tone characteristic determiner 15B changes from the tone characteristic signal TON to the no tone characteristic signal TOFF.

  When there is an output TEND from the signal type determiner 16D, the filter state determiner 31 performs the determination of the above-described conditions 1 to 4 as in the third embodiment, and the coefficient controller according to the determination result. The coefficient control signal ADP_F, ADP_STP, or ADP_RST is output to 17B, and the coefficient update of the adaptive filter 18 is controlled. Except for such control, the adaptive filter 18 follows the coefficient update control of the double talk detector 14B.

  Hereinafter, the operation at the time of transfer will be described. However, unlike the above-described embodiment, in the fifth embodiment, it is not necessary to finely divide the fluctuation of the response characteristic of the hybrid circuit and the convergence state of the adaptive filter. So, I will explain it together.

  Some calls are made before transfer, and the adaptive filter 18 is in the middle of convergence due to the voice. At this time, a far-end speaker (not shown) requests connection to a different telephone by inputting a DTMF signal or the like.

  The DTMF signal input to the reception input terminal Rin is input to the tone determination unit 15B, passes through the reception output terminal Rout as it is, and is input to the D / A converter 7.

  At this time, the tone property determiner 15B detects the tone property of the input signal, but the signal type determiner 16D outputs nothing at this time. Therefore, the filter state determination unit 31 does nothing, and the adaptive filter 18 continues the coefficient update and removes the tone-derived echo. At this time, although the coefficient of the adaptive filter 18 is different from the characteristic of the echo path as described above, the coefficient is updated to a coefficient that can erase the tone well. Therefore, the output of the adder 13 becomes small and ERLE has a large value.

  Next, when the tone signal ends, the signal type determiner 16D outputs the tone end signal TEND to the filter state determiner 31.

  Receiving this signal TEND, the filter state determination unit 31 compares the previously held echo cancellation amount ACANC (n) with the echo cancellation amount ACANC (n + 1) newly output from the echo cancellation amount calculator 32. In this case, assuming that the characteristics of the hybrid circuit 8-1 are the same as those of the hybrid circuit 8-2, at the time when the signal TEND is input regardless of the convergence state before and after the tone end signal TEND The coefficients of the filter 18 are conveniently disturbed to remove only the tone signal, and Condition 2, Condition 3 or Condition 4 is satisfied depending on the amount of echo cancellation in the normal audio signal thereafter. At this time, based on Condition 3 or Condition 4, re-updating of the double talk detector 14B and the adaptive filter 18 is executed as in the third embodiment.

  Further, the condition 1 is satisfied when the coefficient is not significantly disturbed by the DTMF signal before and after the input timing of the signal TEND, and because the echo can be sufficiently removed even in the subsequent audio signal. This occurs only in the extension transfer case C in which the convergence of the prior adaptive filter 18 is sufficient and the characteristics of the hybrid circuit do not change. At this time, the adaptive filter 18 preferably proceeds to update the coefficient as it is, and since the coefficient controller 17B outputs nothing, the adaptive filter 18 and the double-talk detector 14B perform normal echo removal operations as desired. The echo removal can proceed without deterioration of the echo before and after the tone signal is generated.

(E-3) Effect of Fifth Embodiment According to the fifth embodiment, the disturbance of the tone signal cannot be prevented in advance, but the adaptive filter coefficient is reset in most cases when a transfer occurs. The double-talk detector is reset so as not to hinder the renewal coefficient update, and re-convergence is possible regardless of the initial delay of the echo path. It is possible to provide an echo canceller with a small hardware scale because there is no need to prepare a plurality of band rejection filters in advance by returning and starting reconvergence.

(F) Other Embodiments The echo canceller may be configured by combining those that can be combined among the technical ideas of the above embodiments.

  In the third and fourth embodiments, the echo cancellation amount calculator that calculates the square ratio of the input before and after the adder is used to calculate the echo cancellation amount. However, the output signal of the adder and the reception input terminal Rin are used. Or the output signal of the adder and the input signal of the reception output terminal Rout may be used. In this case, the attenuation amount of the echo path itself may be added to Δ1 and Δ2 described in the third and fourth embodiments.

  In the fourth embodiment, the signal component determination unit 40 uses a method that uses FFT to detect an arbitrary tone frequency. However, any method other than FFT can be used as long as it can detect an arbitrary single tone. You may use, It is not limited to the method of utilizing FFT.

  Furthermore, in the second to fourth embodiments, the one having two band rejection filters has been shown, but at least the band rejection filter may be provided in the input stage of the adaptive filter.

  Furthermore, in the description of each of the above embodiments, each component has been described as an image configured with hardware, but it is needless to say that some components may be realized by software.

  In the first and second embodiments, the echo canceller is mounted on the VoIP terminal, and in the third to fifth embodiments, the application example in which the echo canceller is mounted on the private branch exchange is shown. Of course, the apparatus on which the canceller is mounted is not limited to these.

  12, 12A, 12B, 12C, 12D ... echo canceller, 13 ... adder, 14, 14B ... double talk detector, 15, 15B ... tone property determiner, 16, 16A, 16B, 16D ... signal type determiner, 17 , 17B ... coefficient controller, 18 ... adaptive filter, 21, 22 ... band rejection filter, 31 ... filter state determiner, 32 ... echo cancellation amount calculator, 40 ... signal component determiner.

Claims (3)

In an echo canceller that uses an adaptive filter to generate pseudo echoes and removes echoes caused by hybrid circuits,
Tone type determination means for determining whether the far-end input signal is a predetermined type of tone signal;
When the tone type determining means determines that the far-end input signal is a predetermined type of tone signal, the frequency component of the determined predetermined-type tone signal is removed from the far-end input signal to the adaptive filter. An echo canceller comprising: a first band rejection filter provided at an input stage of the adaptive filter. A double-talk detecting means for detecting a double-talk state of the far-end speaker and the near-end speaker;
The echo canceller according to claim 1, wherein the double-talk detection unit receives a near-end input signal that has passed through the first band rejection filter. Adding means for subtracting the pseudo echo signal from the adaptive filter from the near-end input signal;
When the tone type determining means determines that the far-end input signal is a predetermined type of tone signal, the frequency component of the determined predetermined type of tone signal is removed from the near-end input signal to the adding means. The echo canceller according to claim 1, further comprising a second band rejection filter provided at an input stage of the adding means.

Images