JP2009212945A - Echo canceler - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To obtain an echo canceler for achieving sufficient echo processing performance with a minimum arithmetic load, completely without degrading reception voice quality. <P>SOLUTION: A pair of speakers 101a, 101b for outputting stereo reception signals and a pair of microphones 102a, 102b collecting speaker's voices as transmission input signals are disposed right-left symmetrically. A transmission input signal conversion section 103 generates a transmission input sum signal and a transmission input difference signal from collected transmission input signals, and a reception signal conversion section 104 generates a reception input sum signal and a reception input difference signal from reception signals. A first adaptive filter 105 generates a pseudo echo sum signal by filtering the reception input sum signal, and a first subtractor 107 cancels the pseudo echo sum signal from the transmission input sum signal. Similarly, a second adaptive filter section 106 generates a pseudo echo difference signal by filtering the reception input difference signal, and a second subtractor 108 cancels the pseudo echo difference signal form the transmission input difference signal. <P>COPYRIGHT: (C)2009,JPO&INPIT

Description

  The present invention relates to an echo canceller that cancels acoustic echoes mixed in a stereo audio communication system.

  In general, an echo canceller used for processing acoustic echo generated when a microphone picks up the sound of a speaker that outputs a reception signal and superimposes it on a transmission signal, and transmits an acoustic path from the speaker to the microphone. The impulse response representing the characteristic is estimated by an adaptive filter, and the acoustic echo is removed by subtracting the pseudo echo signal generated from this and the received signal from the signal picked up by the microphone. In the case of a communication system in which a telephone call is made using stereo sound, a plurality of speakers (for example, two) are usually provided. In order to cancel the acoustic echo generated in such a communication system, the conventional echo canceller needs to estimate the impulse responses of all combinations of speakers and microphones (for example, four types).

  However, if there is a strong correlation between the signals of each channel of the received voice signal received in stereo, the other party also uses the same stereo voice communication system to pick up the monaural sound emitted by the speaker using a stereo microphone and collect the collected sound. When transmitting and receiving each other, there is a problem that the impulse response cannot be estimated correctly.

  In order to solve this problem, for example, Patent Literature 1 discloses an acoustic echo cancellation technique for a stereo audio communication system having two speakers and two microphones for the left channel and the right channel. In order to cancel the acoustic echo input to one of the two microphones, each impulse response from each speaker detected by this microphone is estimated using a separate adaptive filter. At this time, nonlinear conversion is performed on the left and right channel signals of the received voice, thereby reducing the correlation between the left and right channel signals and improving the estimation accuracy of the impulse response.

JP-A-10-190848 (paragraph numbers 0036 to 0043)

  Since the conventional acoustic echo cancellation technique is configured as described above, the sound is distorted due to the nonlinear conversion of the received voice signal, and the sound quality is deteriorated. Here, if the auditory distortion of the received voice is to be suppressed, the degree of non-linearization is limited, so that the accuracy of acoustic echo cancellation processing is lowered. Thus, there is a problem that the acoustic echo cancellation processing and the sound quality of the received voice are in a trade-off relationship, and the maximum quality cannot be obtained at the same time in both cases. Therefore, conventionally, it has been necessary to find an appropriate compromise between the performance of echo cancellation and the sound quality of the received voice.

  In addition, with respect to all combinations of a plurality of speakers and microphones, each impulse response is simultaneously estimated using the same number of adaptive filters as the number of combinations, and thus there is a problem in that a large calculation load is generated.

  The present invention has been made to solve the above-described problems, and an object thereof is to obtain an echo canceling apparatus that realizes sufficient echo processing performance with a minimum calculation load without any deterioration in received sound quality. To do.

  The echo canceller according to the present invention provides a pair of speakers, a pair of microphones, and a first conversion for converting each collected signal collected by the pair of microphones into a sum signal and a difference signal of each collected signal. A second conversion unit that converts each received signal output as a reproduced sound from a pair of speakers into a sum signal and a difference signal of each received signal, and a sum signal obtained by the second conversion unit. Obtained by a first adaptive filter unit that generates a pseudo-echo sum signal obtained by filtering and estimating an echo included in the sum signal of each collected signal collected by a pair of microphones, and a second conversion unit A second adaptive filter unit that filters the difference signal and generates a pseudo echo difference signal that estimates an echo included in the difference signal of each collected signal collected by the pair of microphones, and a first conversion unit Generate a residual signal by subtracting the pseudo echo sum signal from the resulting sum signal Obtained by a first subtracting unit, a second subtracting unit that generates a residual signal obtained by subtracting the pseudo echo difference signal from the difference signal obtained by the first converting unit, and the first and second subtracting units. And a third converter for converting each residual signal into each collected sound signal from which echoes have been eliminated.

  According to this invention, a pair of speakers and a pair of microphones are provided, and each collected sound signal picked up by the pair of microphones is converted into a sum signal and a difference signal, and also outputted as reproduced sound by the pair of speakers. A pseudo echo sum signal obtained by converting each received signal into a sum signal and a difference signal, filtering the sum signal of each received signal, and estimating an echo contained in the sum signal of each collected signal collected by a pair of microphones Is generated by an adaptive filter, and a pseudo-echo difference signal is generated by an adaptive filter that estimates the echo contained in the difference signal of each collected signal collected by a pair of microphones by filtering the difference signal of each received signal. The residual signal is generated by subtracting the pseudo echo sum signal from the sum signal of each collected sound signal, and the residual signal is generated by subtracting the pseudo echo difference signal from the difference signal of each collected sound signal. Echo signal Since that is configured to convert each collected sound signal has been erased, without reducing any reception quality, it is possible to obtain the echo canceller to realize with minimal computational load sufficient echo processing performance.

Embodiment 1 FIG.
FIG. 1 is a block diagram showing the overall configuration of an echo canceling apparatus according to Embodiment 1 of the present invention. The echo canceller shown in FIG. 1 is connected to, for example, a communication device that transmits and receives a voice signal for a call with an external device, picks up surrounding sound with a microphone, and outputs the sound to the communication device. The present invention is applied to a sound input / output device that outputs a signal received from a device to a speaker and generates sound from the speaker.

  The echo canceller shown in FIG. 1 is applied to a stereo audio input / output device that uses a pair of speakers 101a and 101b that output a received signal as a voice and a pair of microphones 102a and 102b that collect a speaker's voice. Transmission that generates a transmission input sum signal (sum signal of each sound collection signal) and a transmission input difference signal (difference signal of each sound collection signal) from the transmission input signals (sound collection signals) collected by the microphones 102a and 102b An input signal converter (first converter) 103, a received signal converter (first signal) that generates a received input sum signal (sum signal of each received signal) and a received input difference signal (difference signal of each received signal) from the received signal. 2 conversion unit) 104, first or second adaptive filter 105, 106 for generating a pseudo echo sum signal or a pseudo echo difference signal by filtering a received input sum signal or a received input difference signal, a transmission input sum The pseudo-echo sum signal or pseudo-echo difference signal is subtracted from the signal or transmission input difference signal to generate a transmission output sum signal (sum signal component of the residual signal) or a transmission output difference signal (difference signal component of the residual signal). Transmission output signal conversion unit (third conversion unit) that generates a transmission output signal (each collected sound signal from which echo has been canceled) from the first or second subtracters 107 and 108, the transmission output sum signal, and the transmission output difference signal 109.

The speaker 101a shown in FIG. 1 outputs the received signal r _L (n) received from the outside, and the speaker 101b outputs the received signal r _R (n), so that the received voice is reproduced in stereo. In the following, it is assumed that the subscript n represents a discrete time. The subscripts L and R represent the L channel and the R channel, respectively.

The microphones 102a and 102b pick up the voices of the speakers in the installed rooms, convert them into transmission input signals x _L (n) and x _R (n), and output them.

  FIG. 2 is an explanatory diagram showing the positional relationship between the speaker and the microphone of the echo canceling apparatus according to Embodiment 1 of the present invention. In FIG. 2, the same or corresponding parts as those in FIG. Speakers 101a and 101b and microphones 102a and 102b shown in FIG. 2 are installed in the same room and used for transmitting and receiving audio. At this time, as shown in FIG. 2, the pair of speakers 101a and 101b and the pair of microphones 102a and 102b are arranged such that the left and right speakers 101a and 101b and the left and right microphones 102a and 102b Arranged symmetrically. The arrangement position may be other than that shown in FIG. It should be noted that the arrangement is constrained under such conditions only in the speakers 101a and 101b and the microphones 102a and 102b, and the positions of other parts and speakers are arbitrary.

As shown in FIG. 2, since the pair of speakers and the pair of microphones are installed symmetrically, it can be considered that the impulse response from the speaker to the microphone is also approximately symmetrical. In the following description, an impulse response representing the transfer characteristic of the acoustic path from the left speaker 101a to the left microphone 102a and an impulse response representing the transfer characteristic of the acoustic path from the right speaker 101b to the right microphone 102b are respectively represented by F _D. It expresses. Further, an impulse response representing the transfer characteristic of the acoustic path from the left speaker 101a to the right microphone 102b and an impulse response representing the transfer characteristic of the acoustic path from the right speaker 101b to the left microphone 102a are respectively expressed as F _X. To express. These F _D and F _X are numerical sequences expressing the impulse response waveform in discrete time.

The microphones 102a and 102b pick up the voice of the speaker, but the voice of the speaker 101a and 101b is also picked up along with the voice of the speaker. Therefore, the received voice is mixed as an acoustic echo in the transmission input signals x _L (n) and x _R (n). If the speaker speech components included in the transmission input signals x _L (n) and x _R (n) are s _L (n) and s _R (n), the transmission input signals x _L (n) and x _R (n ) Can be represented by the following formula (1).
x _L (n) = s _L (n) + F _D ^* r _L (n) + F _X ^* r _R (n) (1)
x _R (n) = s _R (n) + F _X ^* r _L (n) + F _D ^* r _R (n)
Here, ^* is an abbreviation that represents a convolution operation (the same applies hereinafter).

The transmission input signal conversion unit 103 uses the transmission input signal x _L (n) collected by the microphone 102a and the transmission input signal x _R (n) collected by the microphone 102b according to the following equation (2) as the transmission input sum signal x. _S (n) and the transmission input difference signal x _D (n).
x _S (n) = x _L (n) + x _R (n) (2)
_{x D (n) = x L} (n) -x R (n)

Here, the sum signal s _S (n) and the difference signal s _D (n) of the speaker speech components s _L (n) and s _R (n) are expressed by the following equation (3).
s _S (n) = s _L (n) + s _R (n) (3)
s _D (n) = s _L (n) −s _R (n)

From the above equations (1), (2), and (3), the transmission input sum signal x _S (n) and the transmission input difference signal x _D (n) are expressed as the following equation (4).
x _S (n) = s _S (n) + (F _D + F _X ) ^* {r _L (n) + r _R (n)} (4)
_{x D (n) = s D} (n) + (F D -F X) * {r L (n) -r R (n)}

The received signal converter 104 converts the received signals r _L (n) and r _R (n) into a received input sum signal r _S (n) and a received input difference signal r _D (n) according to the following equation (5). .
r _S (n) = r _L (n) + r _R (n) (5)
r _D (n) = r _L (n) −r _R (n)

From the above formula (5), the above formula (4) is represented by the following formula (6).
x _S (n) = s _S (n) + (F _D + F _X ) ^* r _S (n) (6)
_{x D (n) = s D} (n) + (F D -F X) * r D (n)

Equation (6) is obtained by subtracting, from the transmission input sum signal x _S (n), a pseudo echo signal obtained by filtering the reception input sum signal r _S (n) with a filter having a characteristic represented by F _D + F _X. This shows that a signal obtained by canceling the acoustic echo, which is a complete form of the sum signal component s _S (n) of the speaker voice, that is, a mixture of the received voice is obtained. Similarly, the difference signal component s _D (n) of the speaker voice has a characteristic represented by F _D -F _X from the transmission input difference signal x _D (n) to the reception input difference signal r _D (n). This can be obtained in its entirety by subtracting out the filtered pseudo echo signal.

The transmission input sum signal x _S (n), the transmission input difference signal x _D (n), the reception input sum signal r _S (n), and the reception input difference signal r _D (n) of the above equation (6) are known. Therefore, F _D + F _X and F _D −F _X may be identified using a general learning identification algorithm.

Based on the above formula (6), the operation of acoustic echo cancellation of the echo cancellation apparatus will be described. When the received input sum signal r _S (n) is input from the received signal conversion unit 104, the first adaptive filter 105 receives the received input sum signal r _S by the first filter coefficient sequence H _S according to the following equation (7). Filter (n) and output a pseudo echo sum signal d _S (n). The pseudo echo sum signal d _S (n) is a signal obtained by estimating the acoustic echo included in the transmission input sum signal x _S (n) by the first adaptive filter 105.
d _S (n) = H _S ^* r _S (n) (7)

The first subtracter 107 receives the transmission input sum signal x _S (n) from the transmission input signal converter 103 and the pseudo echo sum signal d _S (n) from the first adaptive filter 105. Then, according to the following equation (8), the first subtracter 107 removes the pseudo echo sum signal d _S (n) from the transmission input sum signal x _S (n) represented by the above equation (6). The signal u _S (n) is output.
u _S (n) = s _S (n) + (F _D + F _X ) ^* r _S (n) −d _S (n) (8)

Here, from the above equation (8), the sum signal error e _S (n), which is the difference between the pseudo echo sum signal d _S (n) and the sum signal s _S (n) of the speaker voice, is expressed by the following equation (9). Define. The following equation (9) can be developed as shown in the lower part based on the above equation (7).
e _S (n) = (F _D + F _X ) ^* r _S (n) −d _S (n) (9)
_{= (F D + F X -H} S) * r S (n)

The transmission output sum signal u _S (n) is output from the first subtractor 107 to the transmission output signal converter 109 and fed back to the first adaptive filter 105. Then, the first adaptive filter 105 uses the transmission output sum signal u _S (n) and the reception input sum signal r _S (n) input from the reception signal conversion unit 104 to generate a sum signal error e _S (n ) Is successively updated so that the first filter coefficient string H _S is minimized. For updating the first filter coefficient string H _S , a known learning identification algorithm such as an LMS (Least Mean Square) algorithm may be used. From the above equation (8), it can be seen that the first filter coefficient sequence H _S may be finally equivalent to F _D + F _X in order to minimize the sum signal error eS (n).

Similarly, the second adaptive filter 106 filters the received input difference signal r _D (n) output from the received signal conversion unit 104 with the second filter coefficient sequence H _D and is expressed by the following equation (10). The pseudo echo difference signal d _D (n) is output. The pseudo echo difference signal d _D (n) is a signal obtained by estimating the acoustic echo included in the transmission input difference signal x _D (n) by the second adaptive filter 106.
d _D (n) = H _D ^* r _D (n) (10)

Further, the second subtracter 108 transmits the transmission output difference signal u _D (n) obtained by removing the pseudo echo difference signal d _D (n) from the transmission input difference signal x _D (n) represented by the above formula (6). Is output.
u _D (n) = s _D (n) + (F _D −F _X ) ^* r _D (n) −d _D (n) (11)

Further, from the above equation (11), the difference signal error e _D (n) which is the difference between the pseudo echo difference signal d _D (n) and the speaker speech difference signal s _D (n) is defined by the following equation (12). To do. The following equation (12) can be developed as shown in the lower part based on the above equation (10).
e _D (n) = (F _D −F _X ) ^* r _D (n) −d _D (n) (12)
_{= (F D -F X -H D} ) * r D (n)

The transmission output difference signal u _D (n) is output from the second subtractor 108 to the transmission output signal converter 109 and fed back to the second adaptive filter 106. Then, the second adaptive filter 106 uses the transmission output difference signal u _D (n) and the received input difference signal r _D (n) input from the received signal conversion unit 104 to calculate a difference signal error e _D (n). There sequentially updates the second filter coefficient sequence H _D so as to minimize. From the equation (12), a second filter coefficient sequence H _D, in order to minimize the difference signal error e _D (n), finally it is understood that it becomes the F _D -F _X equivalent.

When the transmission output signal conversion unit 109 receives the transmission output sum signal u _S (n) output from the first subtracter 107 and the transmission output difference signal u _D (n) output from the second subtractor 108. The transmission output signals u _L (n) and r _R (n) are generated according to the following equation (13) and output to the outside.
u _L (n) = {u _S (n) + u _D (n)} / 2 (13)
u _R (n) = {u _S (n) −u _D (n)} / 2

If the first filter coefficient string H _S and the second filter coefficient string H _D are completely equivalent to F _D + F _X and F _D −F _X , the above equations (3), (8), From the formula (11) and the formula (13), it can be expressed by the following formula (14). Therefore, the transmission output signal from which the acoustic echo is completely removed is output from the echo canceller.
u _L (n) = s _L (n) (14)
u _R (n) = s _R (n)

  As described above, according to the first embodiment, the reception signal and the transmission input signal are converted into the sum signal and the difference signal, and the reception signal and the transmission input signal have a one-to-one relationship as shown in Equation (6). By doing so, the impulse response from the speaker to the microphone can be correctly estimated without being affected by the correlation between the received signals. As a result, it is not necessary to cause distortion of the received signal due to non-linear transformation in order to reduce the correlation between the received signals as in the prior art, and it is possible to simultaneously realize high quality received sound quality and high echo cancellation performance. it can.

  Furthermore, according to the first embodiment, it is only necessary to provide the first and second adaptive filters for canceling the echo for each of the sum signal and the difference signal, and a combination of two speakers and two microphones as in the conventional case. There is no need to provide a number, ie a total of four adaptive filters. Therefore, the calculation load of the echo canceller can be reduced to almost half of that in the prior art.

Embodiment 2. FIG.
When stereo signals are transmitted / received in stereo audio communication, the amount of information can be saved and the encoding efficiency may be higher when the left and right stereo signals are transmitted / received as they are without being transmitted / received as they are. For example, MPEG Audio Layer III). In the present embodiment, an echo canceller is applied to a stereo audio communication system that transmits and receives audio signals in the form of such sum and difference signals.

FIG. 3 is a block diagram showing the overall configuration of the echo cancellation apparatus according to Embodiment 2 of the present invention. In FIG. 3, the same or equivalent parts as in FIG. The received input sum signal r _S (n) and the received input difference signal r _D (n) received from the outside in the form of a sum signal and a difference signal are input to the received signal conversion unit (fourth conversion unit) 201. The received signal conversion unit 201 converts the received input sum signal r _S (n) and the received input difference signal r _D (n) into the left and right stereo received signals r _L (n), r _R (n) according to the following equation (15). Convert to
r _L (n) = {r _S (n) + r _D (n)} / 2 (15)
r _R (n) = {r _S (n) −r _D (n)} / 2

Since then, the echo canceling apparatus of the second embodiment operates in the same manner as the echo canceling apparatus of the first embodiment, so the description will be simplified. The speakers 101a and 101b arranged symmetrically reproduce the reception signals r _L (n) and r _R (n), which are stereo reception signals output from the reception signal conversion unit 201, as received voices.

The microphones 102a and 102b arranged symmetrically pick up the voice of the speaker. At this time, the transmission input signals x _L (n) and x _R (n) picked up by the microphones 102a and 102b include speaker sounds s _L (n) and s _R (n) and acoustic echoes mixed therein. The transmission input signal converting unit 103 adds or subtracts these transmission input signals x _L (n) and x _R (n) to obtain a transmission input sum signal x _S (n) or a transmission input difference signal x _D (n). Convert.

The first adaptive filter 105 adaptively estimates F _D + F _X , which is an impulse response from the speaker to the microphone, using an LMS algorithm or the like, and the estimated filter coefficient sequence H _S and the received input sum signal r _S (n) To generate a pseudo echo sum signal d _S (n). Then, the first subtracter 107 removes the pseudo echo sum signal d _S (n) from the transmission input sum signal x _S (n) to generate a transmission output sum signal u _S (n).

Similarly, the second adaptive filter 106 is the impulse response from the speaker to the microphone F _D and -F _X adaptively estimated by LMS algorithm or the like, the estimated filter coefficient sequence H _D and the received input differential signal r _D The pseudo echo difference signal d _D (n) is generated by convolving (n). Then, the second subtracter 108 removes the pseudo echo difference signal d _D (n) from the transmission input difference signal x _D (n) to generate a transmission output difference signal u _D (n).

In the first embodiment, the transmission output sum signal u _S (n) and the transmission output difference signal u _D (n) are converted into the left and right transmission output signals u _L (n), r _R (n) and output. In this Embodiment 2, it encodes with arbitrary encoding means with the form of a sum signal and a difference signal, and transmits outside.

  As described above, according to the second embodiment, in addition to the effects described in the first embodiment, the echo canceling apparatus is adapted to a stereo sound communication system that transmits and receives stereo sound in the form of a sum signal and a difference signal. By simplifying the configuration, it is possible to efficiently erase the acoustic echo.

It is a block diagram which shows the whole structure of the echo cancellation apparatus which concerns on Embodiment 1 of this invention. It is explanatory drawing which shows the positional relationship of the speaker and microphone of the echo cancellation apparatus which concerns on Embodiment 1 of this invention. It is a block diagram which shows the whole structure of the echo cancellation apparatus which concerns on Embodiment 2 of this invention.

Explanation of symbols

  101a, 101b Speaker, 102a, 102b Microphone, 103 Transmission input signal converter (first converter), 104 Received signal converter (second converter), 105 First adaptive filter, 106 Second adaptive filter , 107 first subtractor, 108 second subtractor, 109 transmission output signal conversion unit (third conversion unit), 201 reception signal conversion unit (fourth conversion unit).

Claims (5)

A pair of speakers;
A pair of microphones,
A first converter that converts each collected signal collected by the pair of microphones into a sum signal and a difference signal of each collected signal;
A second converter that converts each received signal output as reproduced sound by the pair of speakers into a sum signal and a difference signal of the received signals;
Filtering the sum signal obtained by the second conversion unit to generate a pseudo echo sum signal in which an echo included in the sum signal of each collected signal collected by the pair of microphones is estimated An adaptive filter section;
Filtering the difference signal obtained by the second conversion unit to generate a pseudo echo difference signal in which an echo included in the difference signal of each collected signal collected by the pair of microphones is estimated An adaptive filter section;
A first subtraction unit that generates a residual signal obtained by subtracting the pseudo echo sum signal from the sum signal obtained by the first conversion unit;
A second subtraction unit that generates a residual signal obtained by subtracting the pseudo echo difference signal from the difference signal obtained by the first conversion unit;
An echo canceller comprising: a third conversion unit that converts each residual signal obtained by the first and second subtracting units into each collected sound signal from which echo has been canceled. The second conversion unit converts each reception signal reproduced in the left and right stereo format in the pair of speakers into a sum signal and a difference signal of each reception signal,
2. The third converting unit converts each residual signal obtained by the first and second subtracting units into left and right stereo-type sound pickup signals from which echoes are eliminated. Echo canceller. A pair of speakers;
A pair of microphones,
A first converter that converts each collected signal collected by the pair of microphones into a sum signal and a difference signal of each collected signal;
A fourth converter that converts the received sum signal and difference signal into respective received signals that are output as reproduced sounds by the pair of speakers;
A first adaptive filter unit that filters the received sum signal and generates a pseudo echo sum signal that estimates an echo included in the sum signal of each collected signal collected by the pair of microphones;
A second adaptive filter unit that filters the received difference signal and generates a pseudo echo difference signal that estimates an echo included in the difference signal of each collected sound signal collected by the pair of microphones;
A first subtraction unit that generates a residual signal obtained by subtracting the pseudo echo sum signal from the sum signal obtained by the first conversion unit;
An echo canceller comprising: a second subtraction unit that generates a residual signal obtained by subtracting the pseudo echo difference signal from the difference signal obtained by the first conversion unit.   4. The echo canceller according to claim 3, wherein the fourth conversion unit converts the received sum signal and difference signal into received signals reproduced in a left and right stereo format by a pair of speakers. Each of the pair of speakers is arranged symmetrically,
5. The echo canceller according to claim 1, wherein each of the pair of microphones is disposed symmetrically at a position facing the pair of speakers. 6.

Images