JP2012205123A - Acoustic echo canceller - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an acoustic echo canceller capable of efficiently achieving stable echo cancellation capability.SOLUTION: The acoustic echo canceller includes: an audio signal decoding unit 2 for decoding a stereo audio signal using a stereo parameter generated on the basis of correlation between input left and right stereo audio signals; and an acoustic echo cancellation unit 6, including a stereo adaptive filter 62 which estimates an acoustic echo produced by the decoded stereo audio signal, for eliminating the acoustic echo estimated from the input audio signals. The acoustic echo cancellation unit 6 further includes a stereo echo canceller controller 61 for controlling the estimation operation of the acoustic echo in the stereo adaptive filter 62, on the basis of the stereo parameter.

Description

  Embodiments described herein relate generally to an acoustic echo canceller.

  In information home appliances such as a TV and a car navigation system, a method of decoding a digitized stereo audio signal and listening to it with a stereo speaker is generalized.

  Many stereo audio decoders of the international standard system efficiently perform band compression by generating and using stereo parameters using the correlation between left and right channel signals. For example, the intensity stereo system, which is one of the audio encoding modes defined by MPEG, extracts and encodes a monaural signal and a left / right stereo level ratio as a stereo parameter from an audio signal in a frequency domain. Thus, band compression is efficiently performed.

  In recent years, attention has been paid to a new function that enables a microphone to be connected to such a device and enables voice recognition, telephone / videophone, and the like. For example, functions have been developed in which a microphone is connected to a television and the sound volume is adjusted through the microphone or a videophone is used. In addition, for example, a function has been developed in which a microphone is connected to a car navigation system and the navigation system is controlled by voice.

  On the other hand, if you try to do hands-free conversation or voice recognition during stereo audio playback, the echo of the stereo audio signal wraps around the microphone, causing call failure, voice recognition performance degradation, voice quality degradation, etc. due to acoustic echo There's a problem.

  In order to prevent such obstacles / deterioration, a stereo audio echo canceller is usually used to remove a stereo audio signal that has sneak into the microphone from a signal collected by the microphone.

  A typical stereo acoustic echo canceller estimates the echo path characteristics between a speaker and a microphone using system identification. An adaptive filter is generally used for this system identification.

  However, with a conventional acoustic echo canceller, if there is a correlation between the left and right stereo audio signals that are input to the adaptive filter, the correct stereo echo path characteristics cannot be estimated, and the echo cancellation capability when the correlation changes There has been a problem that the material deteriorates significantly. This problem is particularly noticeable when the stereo audio signal has a single statement that is essentially a monaural signal.

  In order to solve this problem, in order to reduce the correlation between the left and right stereo audio signals, a method of giving independent distortion to each channel has been proposed, but in order to distort the stereo audio signal itself, A big problem of causing degradation of voice quality occurs.

JP 2006-314051 A

  The problem to be solved by the present embodiment is to provide an acoustic echo canceller that can realize an efficient and stable echo cancellation capability.

  The acoustic echo canceller of the present embodiment includes an audio signal decoding unit that decodes the stereo audio signal using a stereo parameter generated based on a correlation between the input left and right stereo audio signals, and the decoded A stereo echo estimator for estimating the acoustic echo caused by the stereo audio signal, and an acoustic echo canceling unit for removing the estimated acoustic echo from the input voice signal. The acoustic echo cancellation unit further includes a stereo echo canceller controller that controls an estimation operation of the acoustic echo in the stereo echo estimator based on the stereo parameter.

The schematic block diagram explaining an example of the structure of the acoustic echo canceller concerning 1st and 2nd embodiment. 6 is a flowchart for explaining a control procedure by a stereo echo canceller controller 61 according to the first embodiment.

  Hereinafter, embodiments will be described with reference to the drawings.

(First embodiment)
First, the configuration of an acoustic echo canceller according to the first embodiment of the present invention will be described with reference to FIG. FIG. 1 is a schematic block diagram for explaining an example of the configuration of an acoustic echo canceller according to the first and second embodiments of the present invention. FIG. 1 also shows peripheral devices used in connection with an acoustic echo canceller.

  As shown in FIG. 1, an acoustic echo canceller according to the first embodiment of the present invention receives a stereo audio signal from a tuner or the like and encodes it, and decodes the encoded signal. An audio signal decoding unit 2 that outputs a digital audio signal and an acoustic echo canceling unit 6 that removes an acoustic echo from a voice signal input from a microphone 5 or the like are provided.

  The audio signal decoding unit 2 includes a data decomposer 21 that divides the encoded signal into a plurality of bands, and reproduces the left and right stereo audio signals by decoding the encoded signals for the allocated bands. A stereo signal decoding unit 22 and a band synthesizer 23 for synthesizing the reproduced stereo audio signal of the entire band are provided.

The stereo signal decoding unit 22 is composed of the same number of stereo signal decoders 22 ₀ , 22 ₁ ,..., 22 _T−1 as the number T of bands divided by the data decomposer 21. For example, if the signal coded by the data decomposer 21 is divided into 16 bands, 16 stereo signal decoder ₂₂ 0 is the stereo signal decoding section _22, 22 1, ..., _{22 15} are arranged The Each of the stereo signal decoders 22 ₀ , 22 ₁ ,..., 22 _T−1 includes a band division monaural audio decoder 24 and a correlation generator 25.

  On the other hand, the acoustic echo canceling unit 6 estimates the echo path characteristics between the speakers 3 and 4 and the microphone 5 from the left and right stereo audio signals, and generates a pseudo echo signal, as a stereo adaptive filter 62 as a stereo echo estimator. And a monaural adaptive filter 63 that estimates an echo path characteristic between the speakers 3 and 4 and the microphone 5 based on a monaural signal that is one of the left and right stereo audio signals. One of the left and right stereo audio signals is input to the monaural adaptive filter 63 via the changeover switch 67. The stereo adaptive filter 62 and the monaural adaptive filter 63 are connected by a stereo echo path estimation controller 64.

  The pseudo echo signal estimated and generated by the stereo adaptive filter 62 is subtracted from the audio signal input from the microphone 5 by the adder 65. The audio signal from which the pseudo echo signal is reduced and the echo is removed is output to an external device and simultaneously fed back to the stereo adaptive filter 62 to be used for learning for estimating the echo path characteristic.

  Further, the pseudo echo signal estimated and generated by the monaural adaptive filter 63 is subtracted from the sound signal input from the microphone 5 by the adder 66. The audio signal from which the pseudo echo signal generated by the monaural adaptive filter 63 is reduced is fed back to the monaural adaptive filter 63 and used for learning to estimate the echo path characteristic.

  The acoustic echo canceling unit 6 includes a stereo echo canceller controller 61 that controls the entire acoustic echo canceling unit 6. The stereo echo cancellation unit 6 controls a coefficient used for estimation of echo path characteristics in the stereo adaptive filter 62 via the stereo echo path estimation controller 64. Further, the stereo echo canceller controller 61 controls the changeover switch 67 to select / switch the signal input to the monaural adaptive filter 63.

  The receiver 1 receives a stereo audio signal from a tuner or the like, performs band compression, encodes it, and outputs it to the audio signal decoding unit 2.

  For example, in a recently standardized audio decoding method for stereo, a DCT (Discrete Cosine Transform) or a band division filter is used to efficiently divide a stereo audio signal into a plurality of band components. . In that case, the stereo audio signal often has a correlation between the channels. Therefore, after performing orthogonalization processing as described below for each frame section i and band j to remove the correlation between channels, a monaural encoding method is applied to each channel.

For example, for the i-th frame in the j-th band, the left and right channel stereo audio signals are X _Lij (z) and X _Rij (z), respectively, and the inter-channel transfer function based on the inter-channel correlation is G _RRij (z). , _GRLij (z), _GLRij (z), and _GLLij (z), the tilde X _Mij (z) and the tilde X _Sij (z) are the sum and difference signals of the stereo audio signals of the left and right channels, respectively. Can be represented by the following formula (1).

For example, when there is a relationship like the following formula (2) between the left channel signal and the right channel signal,

(E _Lij (z) represents a non-correlated component between channels.)
A matrix of conjugate complex numbers of the transfer function between channels on the right side of Expression (1) is expressed as Expression (3) below.

When Expression (2) and Expression (3) are substituted into Expression (1), Expression (4) is obtained.

That is, tilde X _Mij (z) and tilde X _Sij (z) can be separated into a correlated component and a non-correlated component. At this time, if the non-correlation component E _Lij (z) can be ignored, the tilde X _Sij (z) becomes zero. Therefore, the correlation information between the tilde X _Mij (z) and the channel is encoded, and the audio signal decoding unit 2 It is sufficient to output to.

In the intensity stereo system of MPEG1 or MPEG2, which is a more specific implementation example, the stereo audio signals X _Rij (z) and X _Lij (z) are assumed as shown in equations (5) and (6), respectively.

In Expressions (5) and (6), l _Rij and l _Lij represent signal level values. Then, the matrix of conjugate complex numbers of the transfer function between channels on the right side of Expression (1) is expressed as Expression (7) below.

When Expression (5), Expression (6), and Expression (7) are inserted into Expression (1), the relationship of Expression (8) is obtained.

That is, based on the relationship of Expression (8), the stereo audio signal is orthogonalized to encode the left and right channel sum signal tilde X _Mij (z) and the level information l _Rij and l _Lij . Since these encoding methods perform processing so that the uncorrelated component becomes zero, the main component of the stereo audio signal and the inter-channel correlation information (transmission) for generating the correlation between the left and right channels are encoded. Function) or an approximate level value.

The encoded signal is input to the data decomposer 21 of the audio signal decoding unit 2. The data decomposer 21 divides the received signal into a preset number T (for example, 16) of bands. By this division processing, specifically, for the j-th band of the i-th frame, the stereo parameter signal which is the main component encoding code I _Mij of the stereo audio signal and the inter-channel correlation information encoding code. p _ij is divided.

The encoded code I _Mij as the main component of the stereo audio signal is input to the band division monaural audio decoder 24 of the stereo signal decoder 22 _j corresponding to each band and is decoded. Further, the stereo parameter signal p _ij is input to the correlation generator 25 of the stereo signal decoder 22 _j corresponding to each band.

The encoded code I _Mij input to the band division monaural audio decoder 24 is decoded to become the audio main signal X _Mij (z ′) of each band. The audio main signal X _Mij (z ′) which is a monaural signal is output to the correlation generator 25 arranged in the same stereo signal decoder 22 _j .

The correlation generator 25 generates inter-channel transfer functions G _Rij (z ′) and G _Lij (z ′) based on the stereo parameter signal p _ij input from the data decomposer 21. Using the inter-channel transfer functions G _Rij (z ′), G _Lij (z ′) and the audio main signal X _Mij (z ′) input from the band division monaural audio decoder 24, the following equation (9), An arithmetic processing based on Expression (10) is performed to generate left and right stereo audio signals X _Lij (z ′) and X _Rij (z ′) for each partial band.

The left and right stereo audio signals X _Lij (z ′) and X _Rij (z ′) generated by the stereo signal decoder 22 _j for each partial band are output to the band synthesizer 23. The band synthesizer 23, using a function bold F to recover the signals of all the bands from the signal of the partial band, all the stereo decoder 22 _0, 22 1, _..., stereo audio signal is inputted from the 22 _T-1 , And are synthesized according to Equation (11) and Equation (12).

The synthesized right stereo audio signal tilde X _Ri (z) is output to the right speaker 3 and reproduced. Similarly, the left stereo audio signal tilde X _Li (z) is output to the left speaker 4 and reproduced.

The left and right stereo audio signals tilde X _Ri (z) and X _Li (z) synthesized by the band synthesizer 23 are also output to the stereo adaptive filter 62 of the acoustic echo canceling unit 6. The stereo adaptive filter 62 estimates the echo path characteristic (pseudo stereo echo path characteristic) between the speakers 3 and 4 and the microphone 5 using the system identification, and generates the pseudo echo signal hat Y _i (z).

If the stereo echo path characteristics between the left and right speakers and the microphone are H _Li (z) and H _Ri (z), and the synthesized signal of the transmitted voice and room noise is N _i (z), the i th The audio signal Y _i (z) input to the microphone in the frame can be expressed as in Expression (13).

On the other hand, the adaptive filter estimates the stereo echo path characteristics between the left and right speakers and the microphone, and holds them internally as the left and right pseudo stereo echo path characteristics hat H _Li (z) and hat H _Ri (z). These and the tilde X _Li (z) and tilde X _Ri (z), which are stereo audio signals output from the left and right speakers, are used to generate a pseudo echo signal represented by Expression (14).

The pseudo stereo echo path characteristic is learned and corrected in the sampling time domain using the stereo audio signal output from the left and right speakers and itself. By subtracting the pseudo echo signal hat Y _i (z) represented by the equation (14) from the audio signal Y _i (z) represented by the equation (13), the stereo audio signal that wraps around the microphone is removed. The output signal E _i (z) is obtained. (See equation (15).)

Here, if the adaptive filter has converged, the pseudo stereo echo path characteristics hat H _Li (z) and hat H _Ri (z) estimated as in the following expressions (16) and (17) are true. It matches the stereo echo path characteristics H _Li (z), H _Ri (z).

Therefore, as shown in Expression (18), the output signal E _i (z) matches the synthesized signal N _i (z) of the transmitted voice and room noise. That is, echoes can be suppressed by converging the adaptive filter.

In order to improve the acoustic echo suppression effect, the pseudo stereo echo path characteristics hat H _Li (z) and hat H _Ri (z) are set to true as shown in the above equations (16) and (17). It is important to match the stereo echo path characteristics H _Li (z) and H _Ri (z). The pseudo stereo echo path characteristic is learned and corrected in the sampling time domain using the stereo audio signal output from the left and right speakers 3 and 4 and itself.

The learning of the pseudo stereo echo path characteristics hat H _Li (z) and hat H _Ri (z) in the stereo adaptive filter 62 is actually performed in the sampling time domain. For example, in the well-known NLMS method (Normalized Least Mean Square), the following equations (19) and (20) are given.

In Expression (19) and Expression (20), the bold hat H _L (k), the bold hat H _R (k), the bold _XL (k), and the bold X _R (k) are N-tap adaptive FIR filters. It is an N-dimensional vector composed of the coefficients of Further, e (k) is a residual signal for echo cancellation, and α is a control parameter called step gain that determines the convergence speed.

The left and right stereo audio signal tildes X _Li (z) and X _Ri (z) synthesized by the band synthesizer 23 are applied to the bold letters X _L (k) and the bold letters X _R (k), respectively, e For (k), the sound signal after the acoustic echo is removed through the adder 65 is applied. The value of α as the step gain is set by the stereo echo path estimation controller 64.

The monaural adaptive filter 63 also estimates an echo path characteristic (pseudo monaural echo path characteristic) hat H _i (z) between one of the speakers 3 and 4 and the microphone 5 using system identification. The pseudo monaural echo path characteristic hat H _i (z) uses one stereo audio signal (monaural signal) selected by the switch 67 among the signals output from the left and right speakers 3 and 4 and the like. In the same manner as the stereo adaptive filter 62, learning is performed and corrected in the sampling time domain.

  The stereo echo path estimation controller 64 controls the value of the step gain α used for learning the pseudo stereo echo path characteristic in the stereo adaptive filter 62 in accordance with an instruction from the stereo echo canceller controller 61. Further, the value of the step gain α ′ used for learning the pseudo monaural echo path characteristic in the monaural adaptive filter 63 is also controlled in accordance with an instruction from the stereo echo canceller controller 61. Further, mutual conversion between the pseudo monaural echo path characteristic and the pseudo stereo echo path characteristic is also performed.

The stereo echo canceller controller 61 is a stereo parameter signal p _ij that is an encoding code of inter-channel correlation information or an inter-channel transfer function G _Rij (z ′), G _Lij (z ′) generated from the stereo parameter signal p _ij. ) And the step gains α and α ′ are controlled based on the monitoring result.

For example, when the inter-channel transfer functions G _Rij (z ′) and G _Lij (z ′) are approximated by only the level values as in the following equations (21) and (22):

The ratio change index μ _i of the transfer function between the left and right channels is obtained by Expression (23).

A method of controlling the value of the step gain α based on the monitoring result of the ratio change index μ _i will be described with reference to FIG. FIG. 2 is a flowchart for explaining a control procedure by the stereo echo canceller controller 61 according to the first embodiment.

First, in step S1, the ratio change index μ _i of the transfer function between the left and right channels in the frame is obtained by Expression (23). Next, μ _i is compared with a preset ratio change index threshold value μ _TH (step S2).

If μ _i falls below the threshold μ _TH in step S2, the stereo echo path estimation controller determines that the correlation between the left and right stereo audio signals is strong and proceeds to step S4, and sets the value of step gain α to 0. The stereo adaptive filter 62 is controlled via 64. At the same time, the monaural adaptive filter 63 is controlled via the stereo echo path estimation controller 64 so as to set the value of the step gain α ′ to 1.

The stereo echo canceller controller 61 sets the value of the step gain α ′ to 1 in step S4 and connects the monaural filter 63 with one of the left and right stereo audio signals having a higher level value. The changeover switch 67 is controlled. Further, the stereo echo path estimation controller 64 uses the pseudo stereo echo path characteristic hat H _Li (z) and the hat H _Ri (z) learned so far to use the pseudo monaural echo path characteristic hat H using a known conversion formula. _i (z) is calculated, and a control instruction is given to set it to the monaural adaptive filter 63.

On the other hand, if μ _i is greater than or equal to the threshold μ _TH in step S2, it is determined that the correlation between the left and right stereo audio signals is weak, the process proceeds to step S3, and the stereo echo path estimation is performed so that the value of step gain α is set to 1. The stereo adaptive filter 62 is controlled via the controller 64. At the same time, the monaural adaptive filter 63 is controlled via the stereo echo path estimation controller 64 so that the value of the step gain α ′ is set to zero.

When steps S3 and S4 are completed, the process proceeds to step S5, advances the frame to be monitored to the next frame, and returns to step S1. In this way, the monitoring of the inter-channel transfer functions G _Rij (z ′) and G _Lij (z ′) generated from the stereo parameter signal p _ij or the stereo parameter signal p _ij transmitted in time series is repeated.

  When there is a strong correlation between the left and right stereo audio signals, acoustic echo cancellation is not possible because the stereo adaptive filter 62 does not converge in the correct direction even if learning of the pseudo stereo echo path characteristics is continued. This has caused a problem that the effect is remarkably reduced. However, in the acoustic echo canceller of this embodiment, when there is a strong correlation between the left and right stereo audio signals, the learning of the pseudo stereo echo path characteristic in the stereo adaptive filter 62 is stopped, and the learning is performed when the correlation becomes weak. In order to resume, a stable acoustic echo cancellation effect can be realized.

  As described above, in the acoustic echo canceller according to the present embodiment, when there is a strong correlation between the left and right stereo audio signals, the pseudo adaptive echo path characteristic learning in the stereo adaptive filter 62 is stopped and erroneous estimation is performed. In order to prevent this, it is possible to prevent a decrease in acoustic echo cancellation capability.

Further, the stereo parameter signal p _ij which is the encoded code of the inter-channel correlation information used in the audio signal decoding unit 2 or the inter-channel transfer function G _Rij (z ′), G generated from the stereo parameter signal p _{ij Since} the strength of correlation between the left and right stereo audio signals is determined using _Lij (z ′), no new correlation detection means is required. Therefore, an acoustic echo canceller having an economically stable cancellation capability can be realized.

  Note that when the correlation between the left and right stereo audio signals is strong, the value of the step gain α does not need to be fixedly controlled to zero, and is controlled to a predetermined value close to zero according to various characteristics. It may be.

(Second Embodiment)
In the above-described embodiment, when one section having a strong correlation between the left and right stereo audio signals appears, the false stereo echo path characteristic learning is stopped to prevent erroneous estimation and prevent the echo canceling ability from deteriorating. It was controlled as such. On the other hand, in the acoustic echo canceller according to the present embodiment, when a plurality of sections having a strong correlation between the left and right stereo audio signals appear, the pseudo stereo echo path using the pseudo monaural echo path characteristics learned during the section is used. The point that the echo cancellation capability is further improved by estimating the characteristics is different.

  The configuration of the acoustic echo canceller of this embodiment is the same as that of the first embodiment described with reference to FIG.

Similarly to the first embodiment, the acoustic echo canceller of the present embodiment also regularly monitors the ratio change index μ _i of the transfer function between the left and right channels in the stereo echo canceller controller 61. Ratio change index in the interval of frame _i-v i +1 from the frame i is below the threshold _{mu TH,} The ratio variation index even in frame i-kappa _i interval -v _i-.kappa.i +1 from frame i-kappa _i is mu _If the stereo parameter signals p _ij and p _i-κij that are the correlation information between channels in these sections are different from each _other and differ, the pseudo stereo echo using the pseudo monaural echo path characteristics learned during these sections is used. Estimate path characteristics.

Specifically, the pseudo monaural echo path characteristic hat H _i (z) of the frame i, the pseudo monaural echo path characteristic hat H _i-κi (z) of the frame i-κ _i , and the expression (24) are defined. , Left and right pseudo stereo echo path characteristic hats H _Li (z) according to the well-known conversion equation (25) using the stereo generation transfer functions G _RLi (z) and G _RLi-κi (z) corresponding to each of them. , And estimate the hat H _Ri (z).

As described above, the estimation of the pseudo stereo echo path characteristic using the two-section pseudo monaural echo path characteristic is performed by the stereo echo path estimation controller 64 based on the control instruction from the stereo echo canceller controller 61. The estimated pseudo stereo echo path characteristics H _Li (z) and hat H _Ri (z) are set in the stereo adaptive filter 62 and used for subsequent pseudo stereo echo path characteristics learning.

  Thus, in the acoustic echo canceller of the present embodiment, when a plurality of sections having a strong correlation between the left and right stereo audio signals appear, using the pseudo monaural echo path characteristic estimated in the same section, Estimate the left and right pseudo stereo echo path characteristics. Thereby, the acoustic echo cancellation capability is further improved. Note that a plurality of sections having a strong correlation that appear may be continuous.

  Although several embodiments of the present invention have been described, these embodiments are presented by way of example and are not intended to limit the scope of the invention. These novel embodiments can be implemented in various other forms, and various omissions, replacements, and changes can be made without departing from the scope of the invention. These embodiments and modifications thereof are included in the scope and gist of the invention, and are included in the invention described in the claims and the equivalents thereof.

2 ... audio signal decoding unit, 6 ... acoustic echo canceling unit, 62 ... stereo adaptive filter,

Claims (5)

An audio signal decoding unit that decodes the stereo audio signal using a stereo parameter generated based on a correlation between the input left and right stereo audio signals;
A stereo echo estimator for estimating an acoustic echo caused by the decoded stereo audio signal; and an acoustic echo canceling unit for removing the estimated acoustic echo from the input speech signal;
An acoustic echo canceller comprising:
The acoustic echo cancellation unit further includes a stereo echo canceller controller that controls an estimation operation of the acoustic echo in the stereo echo estimator based on the stereo parameter.
Acoustic echo canceller.   The acoustic echo canceller according to claim 1, wherein the stereo echo canceller controller controls an estimation operation of the acoustic echo in the stereo echo estimator based on a variation degree of the stereo parameter.   The stereo echo estimator is composed of a multidimensional adaptive filter, and the stereo echo canceller controller controls a convergence speed of coefficients of the multidimensional adaptive filter based on a degree of variation of the stereo parameter. The acoustic echo canceller according to claim 1.   The stereo echo canceller controller suppresses the estimation operation of the acoustic echo in the stereo echo estimator when the degree of variation of the stereo parameter is smaller than a set threshold value. Item 4. The acoustic echo canceller according to any one of Items 3 to 3.   The stereo echo canceller controller has different values for the first stereo parameter in the first section and the second stereo parameter in the first section and the second section, and When the degree of fluctuation of one stereo parameter and the degree of fluctuation of the second stereo parameter are both smaller than a set threshold, the first and second stereo parameters and the estimation in the first section The acoustic echo canceller according to claim 1, wherein a third echo path characteristic is estimated based on the first echo path characteristic obtained and the second echo path characteristic estimated in the second section. .

Images