JP2009164670A - Echo canceler - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To solve a problem wherein, conventionally, a call condition is determined by an input signal at the determination point, thereby the accurate determination of the call condition against disturbance can not be carried out, and an echo canceling effect is not stabilized, on the other hand, an echo canceler waiting for determination changeover for a predetermined period for condition change in order to stabilize the effect can not effectively cancel echo because a reaction is delayed by a time loss in that time. <P>SOLUTION: This echo canceler is provided with: an echo removal means outputting a transmission intermediate signal by subtracting a reception signal filtered by a filter from a transmission input signal collected by a speaker outputting a reception signal as received sound and a microphone; an adaptation filter means inputting the reception signal, the transmission input signal and the transmission intermediate signal, inputting the reception signal and the transmission input signal delayed by first and second delay means in accordance with the determination result of a double talk determination means determining a call condition on a frame basis for a predetermined time, and subjecting the filter to adaptation leaning; and a transfer means transferring a learnt filter coefficient to the echo removal means to rewrite the filter coefficient. <P>COPYRIGHT: (C)2009,JPO&INPIT

Description

  The present invention relates to an echo canceling apparatus for voice communication, and in particular, accurately determines a call state of single talk or double talk and effectively cancels echo.

In general, in an acoustic echo canceling apparatus for voice communication, execution / stopping of adaptive learning of an adaptive filter and operation / stopping of an echo suppressor by amplitude control are performed in the current call state, that is, single talk or double talk. This is based on the determination of whether it is a talk.
In a conventional echo canceling apparatus, this determination is made based on a received input signal or a transmitted input signal input to the apparatus at the time of determination. Such a technique is disclosed in, for example, Japanese Patent Laid-Open No. 10-13310 (Patent Document 1).

In the echo canceller disclosed in Patent Document 1, the echo level is estimated from the current received signal level and the echo gain based on the echo gain calculated in advance, and the echo level and the current residual are calculated. By comparing with the signal level, it is determined whether double talk or not, and adaptive learning control is performed.
However, the double-talk determination based on the instantaneous state quantity is likely to be erroneous under the condition that the disturbance such as the noise and the change of the echo path is large. If the determination result changes many times in a short time due to erroneous determination, the operation of the echo canceller is not stable and the sound quality is significantly lowered.

In order to solve the problem that the determination of the state becomes unstable due to disturbance, a technique of reducing erroneous determination by waiting for switching of the determination for a predetermined time when the state changes is disclosed in, for example, Japanese Patent Application Laid-Open No. 2004-153483. (Patent Document 2).
However, although this method can reduce erroneous determination due to erroneous detection of a change in state, when the state changes truly, delaying the switching of the determination by a predetermined time results in erroneous determination during that time. End up.
That is, in the conventional echo canceller disclosed in Patent Document 2, when a change in the call state is actually observed, a certain amount of time is required to determine its authenticity, and this is a time loss. There is a problem that the response of the apparatus is delayed and the echo cannot be effectively canceled.

JP 10-13310 (paragraph numbers [0016] to [0020]) JP 2004-153483 A (paragraph numbers [0069] to [0071], [0112])

The conventional echo canceller determines whether the talk is single talk or double talk based on the received input signal or the transmitted input signal input to the device at the time of determination. As a result, there is a problem that the echo cancellation effect is not stable.
In addition, in order to stabilize the effect, even if it waits for switching of the determination for a predetermined time with respect to the change of the state, the reaction of the apparatus is delayed due to the time loss during this time, and the echo cannot be effectively erased. There was a problem.

  The present invention has been made in order to solve the above-described problems, and an object of the present invention is to provide an echo canceller that effectively cancels echoes by performing double-talk determination with high accuracy.

The echo canceller according to the present invention is
A speaker that outputs a received input signal as a received voice,
A microphone that picks up the transmitted voice and outputs it as a transmission input signal,
Echo extraction means for extracting a reception input signal filtered by a filter from a transmission input signal picked up by a microphone and outputting a transmission intermediate signal;
A double-talk determination means for inputting a reception input signal, a transmission input signal, and a transmission intermediate signal, and determining a call state every predetermined time frame;
Adaptive filter means for adaptively learning the filter according to the determination result by the double talk determination means,
Transfer means for transferring the filter coefficients learned by the adaptive filter means to the echo removal means at a constant period, and rewriting the filter coefficients;
First delay means for inserting a predetermined delay into the received input signal input to the adaptive filter means;
Second delay means for inserting a predetermined delay into the transmission input signal inputted to the adaptive filter means.

  According to the echo canceller of the present invention, the signal input to the adaptive filter means for generating the filter coefficient for removing the echo is delayed, and the double talk determination means inputs the signal after the time frame to be determined. By making it possible to use the time-series signal data thus determined for the determination, it is possible to determine the call state with high accuracy and effectively perform adaptive learning, thereby improving the echo cancellation effect.

Embodiment 1 FIG.
The echo cancellation apparatus according to the first embodiment of the present invention delays the input signal to the adaptive filter means by a predetermined time, so that it is input after the time-series signal to be determined in the double talk determination. This is an echo canceller that can perform analysis using a time-series signal and thereby perform double-talk determination with high accuracy.

1 is a block diagram showing an echo canceling apparatus according to Embodiment 1 of the present invention. Hereinafter, the details of the present apparatus will be described with reference to FIG.
The received input signal R (n) is reproduced as sound by the speaker 101, and this sound is superimposed on the transmitted voice as an echo and input to the microphone 102. For this reason, the transmission input signal S (n) picked up by the microphone 102 includes transmitted voice and echo. On the other hand, the echo removal means 103 first filters the received input signal R (n) by a FIR (Finite Impulse Response) filter 103a represented by a filter coefficient sequence F _N (k) to provide an echo removal signal D (n ) Is extracted from the transmission input signal S (n) by the adder 103b, and the echo is removed, and the transmission intermediate signal Se (n) is output. This process is expressed by the following Equation 1.

The FIR filter 103a represented by the filter series F _N (k) simulates the spatial transfer characteristic of the echo path, and the echo removal signal D (n) generated thereby is a signal that predicts the incoming echo. It is. The calculation of the filter system sequence F _N (k) is performed using adaptive filter means 105 described later.

  The transmission input signal S (n), the reception input signal R (n), and the transmission intermediate signal Se (n) are input to the double talk determination unit 104. The double talk determination unit 104 receives the transmission input signal S (n) and the reception input signal S (n). The call state is determined using the input signal R (n) and the transmission intermediate signal Se (n). The double-talk determining means 104 includes a means for storing the time-series signal data input in the past for a predetermined L time frame, and the determination of the call state is input in the past for the L time frame. It is assumed that this is performed for a time series signal. That is, the call state determination is performed on the time-series signal data of L time frames before the latest input time-series signal data.

The internal processing of the double talk determining means 104 will be specifically described with reference to FIG. The received input signal R (n), the transmitted input signal S (n), and the transmitted intermediate signal Se (n) input to the double talk determining means 104 are respectively received signal storage unit 201, transmitted input signal storage unit 202, and residual. The signal is input to the signal storage unit 203. Each of these storage means stores each time-series signal data input in the past for a predetermined time.
The first state analysis unit 204 and the second state analysis unit 205 read the time series signal data stored in each of the storage units 201 to 203 at a constant cycle, and analyze the call state.

  The first and second state analysis units 204 and 205 have different analysis time windows, respectively, and based on the time series signal data corresponding to these analysis time windows, the call state is either the single talk on the transmitting side or the receiving side. Judge whether single talk or double talk. For example, a single talk on the transmitting side is regarded as a single talk on the transmitting side when it is detected that the receiving side is silent and the transmitting side is sound using the signal level and the sound detection threshold, and this is regarded as a single talk on the transmitting side. Output as analysis results. As a method for analyzing the single talk and double talk on the receiver side, for example, the method described in Patent Document 1 is a suitable example.

The analysis results by the first state analysis unit 204 and the second state analysis unit 205 are output as the first state analysis result Flg_dt1 (k) and the second state analysis result Flg_dt2 (k), respectively. The analysis results Flg_dt1 (k) and Flg_dt2 (k) are identification signals indicating the call state, and these signals may be expressed numerically. K represents the frame number that has been analyzed.
As described above, the first and second state analysis units 204 and 205 have separate analysis time windows, which are referred to as a first time window and a second time window, respectively. As shown in FIG. 3, the first and second time windows have different ranges, and the first time window has a time window that matches the time frame that is currently determined. Suppose that the second time window starts from this time frame and includes after the L time frame.

By setting such a time window, the first state analysis unit 204 can react quickly to a change in the call state, but has a characteristic that it is vulnerable to instantaneous disturbances. On the other hand, the second state analysis unit 205 is resistant to disturbance and can capture a call state after a time frame to be determined in advance, but has a property of low time resolution.
For this reason, the state determination unit 206 combines these analysis results based on the first state analysis result Flg_dt1 (k) and the second state analysis result Flg_dt2 (k) to obtain a final state determination result Flg_dt ( k) is output. That is, with respect to the first state analysis result Flg_dt1 (k), an error due to disturbance is detected based on the second state analysis result Flg_dt2 (k), and an accurate determination result is obtained by eliminating this error. .

Specifically, the determination is made as follows based on the flowchart of FIG.
First, in step ST1, it is determined whether or not the first state analysis result Flg_dt1 (k) is different from the immediately preceding state determination result Flg_dt (k-1), that is, whether or not the condition of the following equation 2 is satisfied.

If equation (2) holds, the process proceeds to step ST2. If not, the process proceeds to step ST4.
In step ST2, it is determined whether Flg_dt1 (k) matches the second state analysis result Flg_dt2 (k), that is, whether the condition of Expression 3 is satisfied.

If Equation 3 holds, the process proceeds to step ST3. If not, the process proceeds to step ST4.
In step ST3, it is determined that the state has changed, and the determination result Flg_dt (k) is output as Equation 4.

  On the other hand, in step ST4, it is determined that the state has not changed, and the determination result Flg_dt (k) is output as Equation 5.

The determination result Flg_dt (k) output from the double talk determination unit 104 as described above is input to the adaptive filter unit 105. The adaptive filter 105 controls execution and stop of adaptive learning based on the determination result. That is, adaptive learning is performed only when the determination result Flg_dt (k) indicates a single talk on the receiving side.
The adaptive filter means 105 also includes a delayed reception input signal Rd (n) to which a predetermined delay is added by the first delay means 106, and a delay transmission to which the delay is similarly added by the second delay means 107. An input signal Sd (n) is input. Here, the first delay means 106 and the second delay means 107 match the determination result Flg_dt (k) with the time of the delayed transmission input signal Sd (n) and the delayed reception input signal Rd (n). Assume that a delay of L time frames, which is a determination delay, is added in the double talk determination means 104.

The adaptive filter means 105 performs adaptive learning using the input delayed transmission input signal Sd (n) and delayed reception input signal Rd (n), and generates a filter coefficient sequence H _N (k). A well-known algorithm such as an LMS (Least Mean Square) algorithm can be used as an adaptive algorithm for the adaptive filter.
The transfer means 108 reads out the filter system sequence H _N (k) generated by the adaptive learning of the adaptive filter means 105 every predetermined period, transfers it to the echo removal means 103, and transfers it to the filter system sequence F of the FIR filter 103a. Overwrite _N (k).

Further, the transmission intermediate signal Se (n) output by the echo removing means 103 is output from the double delay determination result Flg_dt in the third delay means 109 as in the first delay means 106 and the second delay means 107. A delay transmission intermediate signal Sde (n) is output with a delay of L time frames so that the times of (k) and the transmission intermediate signal Se (n) match.
The delayed transmission intermediate signal Sde (n) is input to the echo suppression means 110, and the echo suppression means 110 performs a suppression process for suppressing residual echo based on the determination result Flg_dt (k) of the double talk determination determination means 104. For this suppression process, for example, a variable attenuator process for controlling the attenuation amount based on the determination result Flg_dt (k) may be used. Specifically, when the determination result Flg_dt (k) indicates a single talk on the receiver side, the first predetermined gain A [dB] is set so as to completely suppress the echo, and when the determination result Flg_dt (k) indicates a double talk, 2 is a variable attenuator that adds a predetermined gain B [dB], otherwise gain 0 [dB], and the relationship between these gains is set as shown in equation (6) .

As is apparent from the above, according to the first embodiment, the signal input to the adaptive filter means for generating the filter for removing the echo is delayed, and the time to be judged in the double talk judging means. By making it possible to use time-series signal data input after the frame for determination, it is possible to accurately determine the call state and effectively perform adaptive learning, thereby improving the echo cancellation effect Play.
Further, by performing the echo suppression process based on the determination result of the call state, it is possible to prevent the transmission voice from being erroneously suppressed in the echo suppression process, and to achieve an effect of realizing a high double talk performance.

Embodiment 2. FIG.
The echo canceling apparatus of the second embodiment makes it possible to easily perform state determination by a single state analysis without providing a plurality of state analyzes with different time windows in the double talk determination unit 104.
FIG. 5 is a block diagram showing the configuration of the double talk judging means 104 of the echo canceller according to the second embodiment of the present invention. Only the single state analysis unit 504 is different from the first embodiment, and the other components are the same as those of the first embodiment.
In FIG. 5, the state analysis unit 504 has a time window similar to that of the second state analysis unit 205 of the first embodiment. That is, the analysis time window of the state analysis unit 504 starts from the time frame that is the object of determination at the present time, and has a region that extends after a predetermined L time frame.

  The state determination unit 505 is based on all the state analysis results including, as a time domain, the time frame that is currently determined in the state analysis result Flg_dtF (k) output from the past to the present by the state analysis unit 504. Determine the state. That is, as shown in FIG. 6, analysis results Flg_dtF (k−L + 1), Flg_dtF (k−L + 2)..., Flg_dtF (k) from the past to the present of the L time frame are used. The state analysis result that occupies the largest number is adopted as the determination result.

  As is apparent from the above, according to the second embodiment, the state analysis is repeated for a signal of a certain time frame while sliding a time window longer than this time frame on the time axis. This makes it possible to suppress misjudgment due to disturbances and perform state judgment with high accuracy only by state analysis using a single time window, and achieve high doubletalk performance with a simple configuration while improving echo cancellation effect. Play.

Embodiment 3 FIG.
In the echo canceller of Embodiment 3, the double-talk determination means 104 expands the time region for determining the single-talk transmission immediately before and immediately after the call state becomes the single-transmission single-talk. In this way, the adaptive filter avoids erroneous learning in a section where there is even a slight amount.

FIG. 7 is a configuration diagram showing the configuration of the double talk determining means 104 in the echo canceling apparatus according to the third embodiment of the present invention. Here, it is assumed that a single state analysis unit is provided as in the second embodiment, but two state analysis units having different time windows may be provided as in the first embodiment. .
Hereinafter, the operation of the echo canceling apparatus according to the third embodiment will be described with reference to FIG. The operations of the reception signal storage unit 701, the transmission input signal storage unit 702, the residual signal storage unit 703, and the state analysis unit 704 are the same as those in the second embodiment and will not be described.

  The state analysis result Flg_dtF (k) output from the state analysis unit 704 is input to the intermediate determination unit 705. Intermediate determination unit 705 performs the same determination process as state determination unit 505 of the second embodiment, and outputs intermediate determination result Flg_dtM (k). The intermediate determination result Flg_dtM (k) is input to the final determination unit 706. The final determination unit 706 further delays the determination by one unit time frame and inputs Flg_dtM (k + 1) in the next frame k + 1. The final determination is performed using Flg_dtM (k-1), Flg_dt (k), and Flg_dt (k + 1) input so far.

  The determination procedure in the final determination unit 706 will be described with reference to the flowchart of FIG. In the flowchart of FIG. 8, “SST” indicates that the determination result is single talk on the transmission side. First, in step ST5, it is determined whether or not the intermediate determination result Flg_dtM (k + 1) is SST, that is, transmission single talk. If it is SST, the process proceeds to step ST7. Migrate to In step ST6, it is determined whether Flg_dtM (k-1) is SST. If it is SST, the process proceeds to step ST7. If not SST, the process proceeds to step ST9.

  In Step ST7, it is determined whether Flg_dtM (k) is other than SST. If it is not SST, the process proceeds to Step ST8, and if it is SST, the process proceeds to Step ST9. In step ST8, unlike the intermediate determination result Flg_dtM (k), the final determination unit 706 determines that the transmission side is single talk, and outputs the final determination result Flg_dt (k) as Flg_dt (k) = SST. . On the other hand, in step ST9, it is determined that the intermediate determination result Flg_dtM (k) is positive, and the final determination result Flg_dt (k) is output as Flg_dt (k) = Flg_dtM (k).

  FIG. 9 and FIG. 10 are conceptual diagrams showing an example of determining the call state in the double talk determining means 104. FIG. In these figures, RST represents received single talk. FIG. 9 shows an example in which the intermediate determination result transits from RST to SST. Here, the final determination result outputs SST one frame before the intermediate determination result. FIG. 10 shows an example in which the intermediate determination result transits from SST to RST. Here, the final determination result outputs SST one frame after the intermediate determination result. As a result, the range to be determined as the transmission single talk is expanded, and the adaptive filter performs adaptive learning on the transmission input signal including the transmission voice even a little, such as the beginning and end of the transmission voice, The learning effect is reduced. Similarly, in echo suppression processing, echo suppression processing is prevented from being erroneously suppressed by echo suppression processing by suppressing echo suppression processing from acting on transmission input signals that contain even a small amount of transmission speech. is doing.

  As is apparent from the above, according to the third embodiment, the adaptive filter means is provided by extending the time interval determined as the transmission single talk at the change point between the transmission single talk state and another state. It is possible to prevent misadaptation due to, and erroneous suppression of transmitted voice by echo suppression means, and to improve the echo cancellation effect while maintaining high double talk performance.

Embodiment 4 FIG.
Since the echo cancellation apparatus according to the first embodiment provides a delay in determining the call state for controlling the adaptive filter, the adaptive filter cannot be used as it is to remove the echo, and a separate filter for removing the echo is provided. There was a need.
The echo canceling apparatus according to the fourth embodiment performs adaptive determination by determining a call state for controlling an adaptive filter without providing a delay, and determining a call state for controlling an echo suppressor by providing a delay. The filter can be used for echo removal as it is, while the determination result determined with high accuracy is used for the echo suppressor so that the transmitted voice is not impaired by the echo suppress means. .

FIG. 11 is a block diagram showing a configuration of an echo canceller according to Embodiment 4 of the present invention. Hereinafter, the operation of the echo canceller according to the fourth embodiment will be described with reference to FIG.
The operations of microphone 1102 and speaker 1101 are the same as those in the first embodiment. The adaptive filter means 1103 filters the received input signal R (n) with a filter to generate a withdrawal signal D (n), and removes the withdrawal signal D (n) from the transmission input signal S (n). The echo is removed by and the transmission intermediate signal Se (n) is output. The adaptive filter means 1103 performs adaptive learning using the transmission intermediate signal Se (n) and the received input signal R (n) so that the echo contained in Se (n) is finally minimized. Update filter coefficients.

The first double talk determining means 1104 determines the call state for the current input signal from the received input signal R (n), the transmitted input signal S (n), and the transmitted intermediate signal Se (n), and the first determination. Outputs the result Flg_dtADP (k). The first determination result Flg_dtADP (k) is input to the adaptive filter unit 1103, and the adaptive filter unit 1103 controls the execution / stopping of the adaptive learning according to the first determination result Flg_dtADP (k).
Further, the second double talk judging means 1105 is connected to a time-series signal input for a certain time in the past from the reception input signal R (n), the transmission input signal S (n), and the transmission intermediate signal Se (n). And the second determination result Flg_dtES (k) is output.
As the configuration and determination method of the second double talk determination means 1105, any of the configurations and determination methods described for the double talk determination means of the first, second, and third embodiments can be used.
Further, the delay means 1106 adds a predetermined delay and outputs the delayed transmission intermediate signal Sde (n) so that the second determination result Flg_dtES (k) matches the time of the transmission intermediate signal Se (n). . Based on the second determination result Flg_dtES (k), the echo suppression means 1107 performs a suppression process for suppressing the residual echo on the delayed transmission intermediate signal Sde (n), and outputs a transmission output signal So (n). .

  As apparent from the above, according to the fourth embodiment, the determination of the call state for controlling the adaptive filter is performed without providing a delay, and the determination of the call state for controlling the echo suppressor is provided with a delay. By doing so, the amount of computation required for the echo removal process is reduced, and erroneous suppression of the transmitted voice by the echo suppress means is prevented, and high double talk performance is realized.

  The echo canceller of the present invention is applied to a voice communication device such as a car navigation system or a mobile phone, and is particularly suitable for application to a hands-free communication device such as a car navigation system.

It is a block diagram which shows the echo cancellation apparatus by Embodiment 1 of this invention. It is a block diagram which shows the structure of a double talk determination means. It is explanatory drawing of each time window of the 1st and 2nd state analysis part, It is a flowchart of the state determination processing operation of the state determination unit, It is a block diagram which shows the structure of the double talk determination means of the echo cancellation apparatus by Embodiment 2 of this invention. FIG. 10 is an explanatory diagram of a time window by a state analysis unit according to the second embodiment. It is a block diagram which shows the structure of the double talk determination means of the echo cancellation apparatus by Embodiment 3 of this invention. 12 is a determination procedure flowchart of a final determination unit according to the third embodiment. It is a conceptual diagram in the case where the intermediate determination result transits from RST to SST in the example of determination of the call state in the double talk determination means. It is a conceptual diagram in the case where the intermediate determination result transits from SST to RST in the example of determination of the call state in the double talk determination means. It is a block diagram which shows the structure of the echo cancellation apparatus by Embodiment 4 of this invention.

Explanation of symbols

  101, 1101; speaker, 102, 1102; microphone, 103; echo removal means, 103a; FIR filter, 103b; adder, 104; double talk determination means, 105, 1103; adaptive filter means, 106; first delay Means 107; second delay means 108; transfer means 109; third delay means 110, 1107; echo suppress means 201, 501, 701; received signal storage section 202, 502, 702; transmission input Signal storage unit, 203, 503, 703; residual signal storage unit, 204; first state analysis unit, 205; second state analysis unit, 206, 505; state determination unit, 504, 704; state analysis unit, 705; intermediate determination unit, 706; final determination unit, 1104; first double talk determination unit, 1105; second double talk determination unit, 1106; delay unit.

Claims (19)

In an acoustic echo canceller for voice communication,
A speaker that outputs a received input signal as a received voice,
A microphone that picks up the transmitted voice and outputs it as a transmission input signal,
Echo extraction means for extracting a reception input signal filtered by a filter from a transmission input signal picked up by a microphone and outputting a transmission output signal;
A double-talk determination means for inputting a reception input signal, a transmission input signal, and a transmission intermediate signal, and determining a call state every predetermined time frame;
Adaptive filter means for inputting a reception input signal and a transmission input signal and adaptively learning the filter according to a determination result by the double talk determination means;
Transfer means for transferring the filter coefficients learned by the adaptive filter means to the echo removal means at a constant period, and rewriting the filter coefficients;
First delay means for inserting a predetermined delay into the received input signal input to the adaptive filter means;
An echo canceller comprising second delay means for inserting a predetermined delay into a transmission input signal inputted to the adaptive filter means.   The double talk determining means is further provided with a storage means for holding a signal input in the past for a predetermined time, and based on the signal held in the storage means, determining the call state for the signal for the predetermined time in the past. The echo canceller according to claim 1, wherein The double-talk determination means includes a first state analysis unit that analyzes a call state with respect to a reception input signal and a transmission input signal through a first time window;
The second time window different from the first time window has a second state analysis unit for analyzing the call state for the received input signal and the transmitted input signal, and the analysis results by the two state analysis units The echo canceling apparatus according to claim 1 or 2, wherein the call state is determined based on the above. The first time window is for a range area equal to the time frame that the double talk determination means currently determines,
The second time window covers a time frame that is currently determined by the double-talk determination unit and a region that includes a predetermined number of frames after the time frame. Echo canceler.   The double-talk determination unit is configured such that the state analysis result by the first state analysis unit is different from the previous determination result of the double-talk determination unit and the state analysis results of the first and second state analysis units match. The analysis results of the first and second state analysis units are output as determination results, and if the above does not apply, the same determination results as the determination results of the immediately preceding double talk determination means are output. The echo canceller according to 3 or 4   The double talk determination means has a state analysis unit that analyzes a call state by a predetermined time window, and the analysis time window of the state analysis unit includes a time frame that is currently determined by the double talk determination means, and It is characterized in that the determination of the call state is made based on all the past analysis results, including the time frame that is the object of determination at the present time within the time window, targeting the area including a predetermined number of frames later. The echo canceller according to claim 1 or 2.   7. The double-talk determination means includes, as a determination result, an analysis result that occupies the largest number of all past analysis results including a time frame that is currently determined in a time window. The echo canceller described in 1.   When the double talk determination means detects a change from another state to the single talk state on the transmission side, it determines that the single talk is on the transmission side from a predetermined time frame before this change point. The echo canceller according to claim 3 or 4, characterized in that   When the change from the single talk state on the transmission side to another state is detected, the double talk determination means determines that the single talk on the transmission side is after a predetermined time frame from this change point. The echo canceller according to claim 3 or 4, wherein Delay means for inserting a predetermined delay into the transmission output signal after echo removal;
The echo canceller according to claim 1 or 2, further comprising echo suppress means for performing echo suppression on the transmission output signal into which the delay is inserted in accordance with a determination result by the double talk determination means. In an acoustic echo canceller for voice communication,
A speaker that outputs a received input signal as a received voice,
A microphone that picks up the transmitted voice and outputs it as a transmission input signal,
Adaptive filter means for subtracting the reception input signal filtered by the filter from the transmission input signal picked up by the microphone and outputting a transmission output signal;
A first double that inputs a reception input signal, a transmission input signal, and a transmission intermediate signal, determines a call state with respect to the latest input signal every predetermined time frame, and updates the filter coefficient of the adaptive filter means based on the determination result Talk determination means,
A second double-talk determining means for inputting a reception input signal, a transmission input signal, and a transmission output signal for a certain period of time and determining a call state with respect to these past input signals;
Echo suppression means for performing echo suppression of the transmission output signal from the adaptive filter means in accordance with the determination result by the second double talk determination means;
An echo canceller comprising delay means for inserting a predetermined delay into a transmission output signal input to the echo suppress means.   12. The echo canceling apparatus according to claim 11, wherein the second double talk determining means further includes storage means for holding a signal input in the past for a predetermined time. The second double-talk determining means includes a first state analysis unit that analyzes a call state with respect to a reception input signal and a transmission input signal through a first time window;
A second state analysis unit that analyzes a call state by a second time window different from the first time window;
The echo canceller according to claim 11 or 12, wherein the state determination is performed based on the analysis results of the two state analysis units. The first time window covers a range area equal to the time frame that the second double talk determination means currently determines,
14. The second time window covers a time frame that is currently determined by the second double-talk determination unit and a region that includes a predetermined number of frames after the time frame. The echo canceller described in 1.   The second double-talk determining means has a state analysis result by the first state analyzing section that is different from the immediately preceding determination result of the second double-talk determining means, and the state analysis of the first and second state analyzing sections. When the results match, the analysis results of the first and second state analysis units are output as the determination results. If the above does not apply, the same determination results as the determination results of the immediately preceding second double talk determination means are output. The echo canceller according to claim 13 or 14, wherein the echo canceller outputs the echo.   The second double talk determination means has a state analysis unit that analyzes the state of a call using a predetermined time window, and the analysis time window of the state analysis unit is currently determined by the second double talk determination means. Determine the state of the call based on all past analysis results, including the time frame and the area that includes a predetermined number of frames after that, and the time frame that is currently being determined is included in the time window. The echo canceller according to claim 11 or 12, wherein:   The second double-talk determining means includes the time frame that is the object of determination at the present time within the time window, and the analysis result that occupies the largest number of all past analysis results is used as the determination result. The echo canceller according to claim 16.   The second double-talk determining means determines that the single-talk on the transmission side is a predetermined time frame before the change point when a change from another state to the single-talk state on the transmission side is detected. The echo canceller according to claim 13 or 14, characterized in that:   The second double-talk determining means, when detecting a change from the single talk state on the transmission side to another state, is a single talk on the transmission side from the change point until a predetermined time frame later. The echo canceller according to claim 13 or 14, wherein the echo canceller is determined.

Images