JP5698110B2 - Multi-channel echo cancellation method, multi-channel echo cancellation apparatus, and program - Google Patents

Description

  The present invention relates to a multi-channel echo cancellation method, a multi-channel echo cancellation apparatus, and a program for canceling acoustic echo in a multi-channel loudspeaker communication system.

  In recent years, multi-channel playback technology has progressed in the direction of expanding the number of channels from stereo to 5.1 channels. However, it is known that the listening area where the sound is reproduced with a high three-dimensional feeling is narrow and is a sweet spot, and the three-dimensional sound is considerably reduced outside the sweet spot. Telepresence TV conferences that use multi-channel playback require playback of a wide listening area that can provide a three-dimensional sound equally to more than 10 participants. As such a multi-channel reproduction technique, research and development of Wave Field Synthesis (hereinafter abbreviated as WFS) has been actively promoted (see Non-Patent Document 1).

  In a two-channel communication conference system of multi-channel loudspeaker type such as a telepresence conference, the received voice is reproduced from the speaker and picked up by a microphone to generate an acoustic echo. Problems arise. In order to realize a comfortable calling environment, an acoustic echo canceller that erases a signal component that circulates acoustically from a speaker to a microphone may be provided.

When the multi-channel communication conference system includes an M (≧ 2) channel reproduction system and an N (≧ 1) channel sound collection system, the acoustic echo canceller performs echo cancellation with the configuration shown in FIG. The operation of the conventional multi-channel echo canceling apparatus 10 will be described with reference to FIG. The conventional multi-channel echo canceller 10 includes M (1 ≦ m ≦ M) speakers 2 ₁ to 2 _M , N (1 ≦ n ≦ N) microphones 3 _{1 to} 3 _N , and a received signal vector conversion unit 100. , N echo replica generators 200 ₁ to 200 _N and an echo canceller 500. The received signal is reproduced as an acoustic signal by the speakers 2 _{1 to} 2 _M , and passes around the microphones 3 _{1 to} 3 _N through an acoustic echo path. The received signal is vectorized by the received signal vector conversion unit 100, and the echo replica generation units 200 ₁ to 200 _N generate an echo replica from the vectorized received signal and the echo path estimation value. The echo cancellation unit 500 performs echo cancellation by subtracting an echo replica from the microphone sound pickup signal. The error signal that is the output of the echo canceller 500 is input to the echo replica generators 200 ₁ to 200 _N. The echo replica generation units 200 ₁ to 200 _N update the echo path estimation value from the received signal and the error signal. When the echo path is accurately estimated, the echo signal and the echo replica signal are substantially equal, and the echo is hardly included in the error signal.

  In order to reduce the amount of calculation, an algorithm that performs echo calculation and filter coefficient update in the frequency domain instead of the time domain has been proposed (see Non-Patent Document 2). FIG. 2 is a block diagram showing a configuration of a multi-channel echo canceller 11 that applies the algorithm of Non-Patent Document 2 to multi-channels. FIG. 3 is a flowchart showing the operation of the conventional multi-channel echo canceller 11. Hereinafter, the operation of the multichannel echo canceller 11 will be described in detail with reference to FIGS. In the following description, k represents time, f represents frequency, and j represents a frame number.

The conventional multi-channel echo canceller 11 includes M (1 ≦ m ≦ M) speakers 2 ₁ to 2 _M , N (1 ≦ n ≦ N) microphones 3 _{1 to} 3 _N , and received signal vector conversion. Unit 110, N echo replica generation units 210 _{1 to} 210 _N , an inverse FFT unit 400, an echo cancellation unit 510, and an FFT unit 600.

The received signal vector conversion unit 110 blocks the M channel received signal x _m (k) for each L samples, converts 1 frame = 2L samples, and converts one frame into the frequency domain by fast Fourier transform. The received signal X _m (f, j) is generated as shown in FIG. Here, L is a natural number, and the frame division number D is a natural number that divides L. In order to simplify and speed up the fast Fourier transform, L is often a power of 2.

The echo replica generation unit 210 _n multiplies the received signal X _m (f, j) by the filter coefficient H _{m, n} (f, j) for each frequency f as shown in Expression (2), thereby receiving the received signal X Filter _m (f, j) and add it for M channels. Thus, the echo replica Y ^ _n (f, j) is obtained (S210).

The inverse FFT unit 400 converts the echo replica Y ^ _n (f, j) into the time domain by inverse fast Fourier transform, and obtains the echo replica y ^ _n (j) as shown in Equation (3) (S400).

Here, 0 _L is an L × L zero matrix, and _IL is an L × L unit matrix.

The echo canceling unit 510 receives an error signal e _n (j) from the N-channel transmission signal y _n (j) and the echo replica y ^ _n (j) collected from the _N microphones 3 _{1 to} 3 _{N in} the time domain. ) Is obtained (S510).

The FFT unit 600 converts the error signal e _n (j) to the frequency domain by fast Fourier transform, and obtains the error signal E _n (f, j) as shown in Equation (4) (S600).

The echo replica generation unit 210 _n uses the error signal E _n (f, j) and the received signal X _m (f, j) to adjust the filter coefficient correction amount dH _{m, n} (f, j) as shown in Equation (5). Ask for.

However, ^* (superscript asterisk) represents a complex conjugate.

Next, the filter coefficient H _{m, n} (f, j) of each channel is updated as shown in Expression (6) (S700).

Here, p (f, j) is obtained by calculating the sum of transmission signal powers for N channels for each frequency component as shown in Expression (7), and the correction amount dH _{m, n} (f, j) is corrected.

However, μ is a step size that takes a value of 0 to 1, δ is a small positive constant for preventing the denominator from becoming 0, and β is a short time in power calculation that takes a value of 0 to 1. This is a smoothing constant for averaging.

A. J. Berkhout, D. de Vries, and P. Vogel, “Acoustic control by wave field synthesis”, Journal of Acoustic Society of America, vol. 93, no. 5, pp.2764-2778 (1993). D. Mansour and A. H. Gray, “Unconstrained Frequency-Domain Adaptive Filter”, IEEE Trans. On Acoust., Speech, Signal Processing, vol. ASSP-30, No. 5, pp. 726-734 (1982).

  However, WFS described in Non-Patent Document 1 requires tens or more microphones and tens or more speakers in order to acquire a sound wave surface at a certain point and re-synthesize at another point. Therefore, when introducing WFS into a two-way video conference, the acoustic path between the speaker and microphone becomes the square of the number of channels, so the number of acoustic paths estimated by the echo canceller increases rapidly with the square of the number of channels. .

  Even if the method of performing echo calculation and filter coefficient update in the frequency domain described in Non-Patent Document 2 is used to reduce the computation amount of the echo canceller, the process of echo replica generation and filter coefficient update is performed in the input channel. It increases by the product of the number and the number of output channels, that is, M × N.

  The present invention has been made in view of the above points, and provides a multi-channel echo canceling apparatus capable of reducing the amount of computation in multi-channel echo canceling processing for canceling acoustic echoes in a multi-channel loudspeaker communication system. With the goal.

In order to solve the above problems, the multi-channel echo canceller of the present invention has M (1 ≦ m ≦ M) speakers and N (3 ≦ n ≦ N) speakers arranged at equal intervals on a straight line. A microphone, a received signal vector conversion unit, an echo replica generation unit, an echo replica space interpolation unit, an inverse FFT unit, an echo cancellation unit, and an FFT unit are provided. k represents time, f represents frequency, j represents frame number, i represents adaptive filter tap number, m represents speaker number, n represents microphone number, and SetN (f) is It is assumed that a set of microphone numbers selected so that the spatial sampling interval becomes wider as the frequency f becomes lower from N microphones arranged at equal intervals in a straight line is assumed. The received signal vector conversion unit converts the M channel received signal x _m (k) output from the speaker into a frequency domain for each channel m, and generates a received signal X _m (f, j). When the reception signal X _m (f, j) is input, the echo replica generation unit taps the microphone n included in SetN (f) for each frequency f from the reception signal X _m (f, j). Echo replica Y ^ _{SetN (f)} (f ₎ spatially thinned using filter coefficients H _{m, n, i} (f, j) having a number I (1 ≦ i ≦ I, I is 1 or more) , J). The echo replica spatial interpolation unit performs spatial interpolation for each frequency f from the spatially thinned echo replica Y ^ _{SetN (f)} (f, j), and obtains the echo replica Y ^ _n (f, j). Generate. The inverse FFT unit converts the echo replica Y ^ _n (f, j) into the time domain and generates an echo replica y ^ _n (j). Echo cancellation unit, the transmission signals of N channels picked up from the microphone _y n (j) and the echo replica _{y ^} n (j), generates an error signal _e n (k). FFT unit, an error signal _e n (k), is converted into the frequency domain to generate an error signal _E n (f, j). When the error signal E _n (f, j) is input, the echo replica generation unit sets Set N (f) for each frequency f from the error signal E _n (f, j) and the received signal X _m (f, j). f) A correction amount dH _{m, n, i} (f, j) is obtained for the microphone n included in f), and a filter coefficient H _{m, n, i} is obtained using the correction amount dH _{m, n, i} (f, j) _. Update _i (f, j).

  According to the multi-channel echo canceller of the present invention, focusing on the spatial shape of the echo signal for each frequency, generating an echo replica using spatial interpolation at a low frequency, generating an echo replica and updating a filter coefficient The number of computations can be reduced, and the amount of computation in the entire echo cancellation processing in a multi-channel loudspeaker communication system can be reduced.

1 is a block diagram showing a configuration of a conventional multi-channel echo canceller 10. The block diagram which shows the structure of the conventional multichannel echo cancellation apparatus 11. FIG. 6 is a flowchart showing the operation of a conventional multi-channel echo canceller 11. The figure which showed the spatial shape for every frequency of an echo signal. The figure which showed the thinning pattern for every frequency. 1 is a block diagram illustrating a configuration of a multichannel echo canceller 20 according to a first embodiment. 3 is a flowchart showing the operation of the multi-channel echo cancellation apparatus 20 according to the first embodiment. The figure which showed the spatial interpolation method of an echo replica. The figure which showed the synthetic | combination method of the error signal. The figure which showed the experimental result of Example 1. FIG. FIG. 4 is a block diagram showing a configuration of a multi-channel echo cancellation apparatus 30 according to the second embodiment. 9 is a flowchart showing the operation of the multi-channel echo cancellation apparatus 30 according to the second embodiment. The figure which showed the experimental result of Example 2. FIG. FIG. 6 is a block diagram showing a configuration of a multi-channel echo cancellation apparatus 40 according to the third embodiment. FIG. 9 is a block diagram showing a configuration of a band division voice switch unit 700 according to the third embodiment. 10 is a flowchart showing the operation of the multi-channel echo cancellation apparatus 40 according to the third embodiment.

  Hereinafter, embodiments of the present invention will be described in detail. In addition, the same number is attached | subjected to the structure part which has the same function, and duplication description is abbreviate | omitted.

[Principle of the present invention]
Prior to the description of the embodiments, the principle of the multi-channel echo cancellation method of the present invention will be described. The present invention focuses on the shape of the echo signal in the spatial direction for each frequency and generates an echo replica using spatial interpolation at a low frequency, thereby suppressing a rapid increase in the amount of echo canceller computation. When the microphones are arranged in a straight line at equal intervals, the spatial shape of the echo signal at each frequency becomes smoother as the frequency is lower. This is because there is a relationship of wavelength λ × frequency f = sound speed c (constant) between the wavelength λ of the sound wave and the frequency f, and the wavelength λ becomes longer as the frequency f is lower. Furthermore, according to the Nyquist theorem, the sampling interval can be much larger than the microphone interval at low frequencies. This means that the spatial sampling interval can be increased at low frequencies, that is, sampling can be thinned out.

  Here, as an example of the WFS sound collection / reproduction system, consider that a linear speaker array and a linear microphone array are operated at a sampling frequency of 16 kHz. According to the Nyquist theorem, the microphone interval, that is, the sampling interval of the spatial wave needs to be λ / 2 or less. In order to sample a wave having a band upper limit of 8 kHz, it is necessary to set the sound speed to 330 [m / s] and the microphone interval to 2 cm. However, the lower the frequency, the smoother the waveform. FIG. 4 shows this situation. The spatial shapes of echo signals of 4 kHz, 2 kHz, 1 kHz, 500 Hz, 250 Hz, and 125 Hz obtained by frequency-resolving one frame of echo signals collected by 33 microphones by fast Fourier transform are plotted. The solid line represents the real number component, and the dotted line represents the imaginary number component. Spatial sampling can be thinned out for smooth waveforms. That is, for low frequencies, echo cancellation processing is performed with some microphones instead of all microphones, and echoes of all microphones are echoed from the echo canceled signals of some microphones without using an adaptive filter for the remaining microphones by spatial interpolation. An erased signal can be determined.

  FIG. 5 shows a thinning pattern SetN (f) of a transmission signal for each frequency f when the number of microphones is 9 and the number of frequency band divisions is 8. The horizontal axis is the microphone position, and the vertical axis is the frequency. The larger the value of f, the higher the frequency. At f = 3, 4, the spatial sampling is decimated by half. At f = 1, 2, spatial sampling is thinned out to 1/4.

The thinned pattern SetN (f) of the transmission signal is obtained from the wavelength λ (f) = c (sound speed) / f of the frequency f. When microphones are arranged at equal intervals, S _N represents the microphone interval, and the maximum natural number Q satisfying S _N × Q ≦ λ (f) / 2 is obtained, and the thinning interval Q (f) is set to be equal to or less than Q. To do. For example, when the number of microphones N = 9 and the thinning interval Q (f) = 4, SetN (f) = {1, 5, 9} can be set. Further, when the thinning interval Q (f) = 2, SetN (f) = {1, 3, 5, 7, 9} can be set.

  Based on the above principle, it is possible to reduce the amount of calculation by thinning out the echo replica generation process and the filter coefficient update process at a low frequency.

  With reference to FIGS. 6 and 7, the operation of the multi-channel echo canceling apparatus 20 according to the first embodiment of the present invention will be described in detail. FIG. 6 is a block diagram showing the configuration of the multi-channel echo cancellation apparatus 20 according to the first embodiment of the present invention. FIG. 7 is a flowchart showing the operation of the multi-channel echo cancellation apparatus 20 according to the first embodiment of the present invention. In this embodiment, the frequency domain adaptation algorithm is “E. Moulines, OA Amrane, and Y. Grenier,“ The Generalized Multidelay Adaptive Filter: Structure and Convergence Analysis ”, IEEE Trans. On SP, vol. 43, no. 1 (1995). "Will be described. In the following description, it is assumed that the adaptive filter is divided into I pieces (I is 1 or more) in the time direction.

In the following, description will be made in the order of procedures actually performed. The multi-channel echo canceling apparatus 20 of this embodiment includes M (1 ≦ m ≦ M) speakers 2 ₁ to 2 _M , N (3 ≦ n ≦ N) microphones 3 _{1 to} 3 _N, and a received signal. A vector conversion unit 110, N echo replica generation units 220 _{1 to} 220 _N , an echo replica space interpolation unit 300, an inverse FFT unit 400, an echo cancellation unit 510, and an FFT unit 600 are provided.

The received signal vector conversion unit 110 blocks the M channel received signal x _m (k) for each L samples, converts 1 frame = 2L samples, and converts one frame into the frequency domain by fast Fourier transform, The received signal X _m (f, j) is generated as in 1) (S110).

For each frequency f, the echo replica generation unit 220 _n receives the received signal X _m (f, j) and the filter coefficient H for n included in SetN (f) for each frequency f as shown in Equation (8). By multiplying _{m, n, i} (f, j), the received signal X _m (f, j) is filtered and added for M channels. Thereby, the echo replica Y ^ _{SetN (f)} (f, j) spatially thinned is obtained (S220).

The echo replica spatial interpolation unit 300 uses, for each frequency f, the spatially thinned echo replica Y ^ _{SetN (f)} (f, j) generated by the echo replica generation unit 210 _n for the frequency f ≦ L + 1. Then, the echo replica Y ^ _n (f, j) is obtained by spatial interpolation (S300). More specifically, the echo replica Y ^ _{n is} obtained by spatially interpolating the echo canceled signal of the channel included in SetN (f) for the channel n skipped by decimation, that is, n not included in SetN (f). Find (f, j). When the microphones are arranged in a straight line at equal intervals, spatial interpolation is performed as shown in Equation (9) using the spatially thinned echo replica Y ^ _{SetN (f)} (f, j) and the sinc function. Is possible.

  In actual calculation, the sinc function needs to be cut off with a finite length. The range in which the sinc function is terminated can be set as in Expression (10) or Expression (11) based on the thinning interval Q (f). In this case, only 4 or 6 echo replicas are required for multiplication with the sinc function.

Subsequently, the echo replica space interpolation unit 300 uses the symmetry regarding the fast Fourier transform result of the real signal for each frequency f with respect to the frequency f> L + 1, and the echo replica Y ^ _n (f , J) is obtained (S300). Here, conj represents taking a complex conjugate.

The inverse FFT unit 400 converts the echo replica Y ^ _n (f, j) into the time domain by inverse fast Fourier transform, and obtains the echo replica y ^ _n (j) as in the above equation (3) (S400).

The echo canceling unit 510 receives an error signal e _n (j) from the N-channel transmission signal y _n (j) and the echo replica y ^ _n (j) collected from the _N microphones 3 _{1 to} 3 _{N in} the time domain. ) Is obtained (S510).

The FFT unit 600 converts the error signal e _n (j) into the frequency domain by fast Fourier transform, and obtains the error signal E _n (f, j) as in the above equation (4) (S600). When the frame division number D is set to D ≧ 2 when the received signal x _m (k) is framed, the error signal E _n (k, j) obtained from the frame number j and the previous frame number The error signal E _n (k, j−1) obtained at j−1 is synthesized through a windowing process and output. FIG. 9 is a diagram showing a composition process when D = 2. Specifically, represents j th frame by obtained error signal _{e n (k 0 + t,} j) a by the equation (13), as _{W H} represents a Hanning window of length 2L / D, after synthesis The error signal e _n ′ (k ₀ + t) can be expressed as in Expression (14).

The echo replica generation unit 220 _n uses the error _coefficient E _n (f, j) and the received signal X _m (f, j) only for n included in SetN (f), as shown in Expression (15). Correction amount dH _{m, n, i} (f, j) is obtained.

Next, for only n included in SetN (f), the filter coefficient H _{m, n, i} (f, j) of each channel is updated as shown in Expression (16) (S710).

Here, p (f, j) is obtained by calculating the sum of transmission signal powers for N channels for each frequency component as shown in Expression (17), and the correction amount dH _{m, n, i} (f, j) are corrected.

Here, μ is a step size that takes a value from 0 to 1, δ is a small positive constant for preventing the denominator from becoming 0, and β is a short power calculation that takes a value from 0 to 1. This is a smoothing constant for taking a time average.

As is apparent from the above, in this embodiment, for n not included in SetN (f), the echo replica y ^ _n (j) and the correction amount dH _{m, n, i} (f, j) are both filter coefficients H. _{Since m, n, i} (f, j) are not calculated, the amount of calculation in the entire multi-channel echo cancellation process can be reduced.

[Experimental result of Example 1]
In order to confirm the effect of this example, a simulation was performed. In a room with a reverberation time of 200 ms, a linear speaker array (33 elements, spacing 2 cm) and a linear microphone array (33 elements, spacing 2 cm) are placed 2 m apart in parallel, and the total echo path impulse response between the speaker and microphone is Generated by simulator. The sampling frequency was set to 16 kHz. When obtaining an echo replica by spatial interpolation, the above equation (10) was used based on the thinning interval Q (f).

  The reception signal was generated by simulating sound collection by 33 microphones using pink noise as a sound source separately. FIG. 10 shows the result of echo cancellation processing according to the configuration of this embodiment. In FIG. 10, for the first, fifth, eleventh, seventeenth, and nineteenth channels among the 33 channels, the levels of the transmission signal and the error signal are plotted with a solid line for the reception signal and a broken line with the error signal. This shows that the echo is erased by the difference between the solid line and the broken line. It can be seen that both echoes are erased well.

[Modification]
In the present embodiment, the case where the frequency domain filter coefficient H _{m, n, i} (f, j) has I taps at any frequency f is described. As a modification of the present embodiment, it is conceivable to reduce the number of taps from the I tap in the high frequency part of the filter coefficient. For example, it is conceivable to set the number of taps to 1 at a frequency exceeding 4 kHz where the human voice component is relatively small. Thereby, it is possible to further reduce the total calculation amount.

In order to further reduce the amount of calculation, the received signal input to the adaptive filter may be thinned out at a low frequency. The reception signal thinning pattern SetM (f) for each frequency f is obtained from the wavelength λ (f) = c / f of the frequency f, similarly to SetN (f). When the speakers are arranged at equal intervals, S _M represents the speaker interval, the maximum natural number Q satisfying S _M × Q ≦ λ (f) / 2 is obtained, and the thinning interval Q _M (f) is set to Q or less. Set.

With reference to FIGS. 11 and 12, the operation of the multi-channel echo canceling apparatus 30 according to the second embodiment of the present invention will be described in detail. FIG. 11 is a block diagram showing a configuration of a multi-channel echo cancellation apparatus 30 according to the second embodiment of the present invention. FIG. 12 is a flowchart showing the operation of the multi-channel echo cancellation apparatus 30 according to the second embodiment of the present invention. The multi-channel echo canceller 30 according to the second embodiment is different from the multi-channel echo canceller 20 according to the first embodiment in that an echo replica generator 230 _n is provided instead of the echo replica generator 220 _n .

The echo replica generation unit 230 _n extracts the received signal X _m (f, j) in which _m is included in SetM (f) for each frequency f with respect to the frequency f ≦ L + 1 (S231). Subsequently, the echo replica generation unit 230 _n sets the received signal X _m (f, j) and _n for frequency f ≦ L + 1 to n included in SetN (f) for each frequency f as shown in Expression (18). By multiplying the filter coefficient H _{m, n, i} (f, j), the received signal X _m (f, j) is filtered and added for M channels. As a result, the echo replica Y ^ _{SetN (f)} (f, j) spatially thinned is obtained (S232).

Further, the echo replica generation unit 230 _n only includes the error signal E _n (f, j) and the received signal X _m (f) when m is included in SetM (f) and n is included in SetN (f). , J), the correction amount dH _{m, n, i} (f, j) of the filter coefficient is obtained as in the above equation (15). Next, only when m is included in SetM (f) and n is included in SetN (f), the filter coefficient H _{m, n, i} (f, j) of each channel is _expressed by the above equation (16). (S720). However, p (f, j) for correcting the correction amount dH _{m, n, i} (f, j) is obtained by calculation as shown in Expression (19).

Here, μ is a step size that takes a value from 0 to 1, δ is a small positive constant for preventing the denominator from becoming 0, and β is a short power calculation that takes a value from 0 to 1. This is a smoothing constant for taking a time average.

As is clear from the above, in this embodiment, the echo replica Y ^ _n (f, j) and the correction amount dH _{m, n, m} only for m and n of m∈SetM (f) and n∈SetN (f) _{. i} (f, j) is calculated and the filter coefficients _{Hm, n, i} (f, j) are updated. Therefore, for m and n other than the above, the echo replica Y ^ _n (f, j) and the correction amount dH _{m, n, i} (f, j) are both filter coefficients H _{m, n, i} (f, j). Is not calculated, the amount of calculation can be further reduced.

  In this embodiment, since the signal on the receiving side is thinned out, the filter coefficient and the echo path characteristic between the specific speaker and the specific microphone do not correspond one-to-one.

[Experimental result of Example 2]
In order to confirm the effect of this example, a simulation was performed with the same settings as in Example 1. FIG. 13 shows the result of echo cancellation processing according to the configuration of this embodiment. In FIG. 13, for the first, fifth, eleventh, seventeenth and nineteenth channels among the 33 channels, the levels of the transmission signal and error signal are plotted with the solid line for the reception signal and the broken line with the error signal. This shows that the echo is erased by the difference between the solid line and the broken line. It can be seen that both echoes are erased well.

  The operation of the multichannel echo canceller 40 according to the third embodiment of the present invention will be described in detail with reference to FIGS. FIG. 14 is a block diagram showing a configuration of a multi-channel echo canceling apparatus 40 according to Embodiment 3 of the present invention. FIG. 16 is a flowchart showing the operation of the multi-channel echo canceling apparatus 40 according to the third embodiment of the present invention. The multi-channel echo canceling apparatus 40 according to the third embodiment is different from the multi-channel echo canceling apparatus 20 according to the first embodiment in that it further includes a band division voice switch unit 800. FIG. 15 is a block diagram showing a configuration of the band division voice switch unit 800. The received signal and the transmitted signal are divided into a low-frequency component and a high-frequency component by the band division voice switch unit 800, and the echo of the high-frequency component is controlled in accordance with the transmission / reception state to adapt the low-frequency component echo. Control by erasing with a filter.

The band division voice switch unit 800 includes a transmission / reception determination unit 810, M reception side high frequency attenuation units 820 _{1 to} 820 _M , and N reception side high frequency attenuation units 830 _{1 to} 830 _N. The reception-side high-frequency attenuation unit 820 _m includes a high-pass filter (hereinafter abbreviated as HPF) 8210, a low-pass filter (hereinafter abbreviated as LPF) 8220, a signal attenuator 8230, and a signal adder 8240. The transmission-side high-frequency attenuation unit 830 _n includes an HPF 8310, an LPF 8320, a signal attenuator 8330, and a signal adder 8340.

The reception-side high-frequency attenuation unit 820 _m applies the HPF 8210 and the LPF 8220 to the reception signal and divides them into a low-frequency signal and a high-frequency signal. The signal attenuator 8230 attenuates the high frequency signal output from the HPF 8210 by a specified amount. The signal adder 8240 adds the signal output from the signal attenuator 8230 and the signal output from the LPF 8220. The transmission side high frequency attenuating unit 820 _n applies the HPF 8310 and the LPF 8320 to the transmission signal and divides the signal into a low frequency signal and a high frequency signal. The signal attenuator 8330 attenuates the high frequency signal output from the HPF 8310 by a specified amount. The signal adder 8340 adds the signal output from the signal attenuator 8330 and the signal output from the LPF 8320.

  The transmission / reception determination unit 810 performs transmission / reception determination using the M channel signal output from the LPF 8220 and the N channel signal output from the LPF 8320 (S810). When it is determined that the call is in the receiving state, only the receiving high frequency signal is attenuated (S820). When it is determined that the transmission state is established, only the high frequency signal on the transmission side is attenuated (S830). The amount of attenuation specified for the signal attenuator 8220 and the signal attenuator 8320 is set in the range of 3 to 40 dB so that the residual echo is at a level that does not matter.

According to the present embodiment, since the echo replica generation unit 220 _m does not use the filter coefficient H _{m, n} (f, j) corresponding to the high-frequency component for echo replica generation, the total amount of computation can be reduced. . However, when the received signal vector conversion unit 110 frames the received signal in order to ensure that no abnormal sound is generated between frames, it is necessary to set the frame division number D to D ≧ 2.

<Program, recording medium>
The various processes described above are not only executed in time series according to the description, but may also be executed in parallel or individually as required by the processing capability of the apparatus that executes the processes. Needless to say, other modifications are possible without departing from the spirit of the present invention.

  Further, when the above-described configuration is realized by a computer, processing contents of functions that each device should have are described by a program. The processing functions are realized on the computer by executing the program on the computer.

  The program describing the processing contents can be recorded on a computer-readable recording medium. As the computer-readable recording medium, for example, any recording medium such as a magnetic recording device, an optical disk, a magneto-optical recording medium, and a semiconductor memory may be used.

  The program is distributed by selling, transferring, or lending a portable recording medium such as a DVD or CD-ROM in which the program is recorded. Furthermore, the program may be distributed by storing the program in a storage device of the server computer and transferring the program from the server computer to another computer via a network.

  A computer that executes such a program first stores, for example, a program recorded on a portable recording medium or a program transferred from a server computer in its own storage device. When executing the process, the computer reads a program stored in its own recording medium and executes a process according to the read program. As another execution form of the program, the computer may directly read the program from a portable recording medium and execute processing according to the program, and the program is transferred from the server computer to the computer. Each time, the processing according to the received program may be executed sequentially. Also, the program is not transferred from the server computer to the computer, and the above-described processing is executed by a so-called ASP (Application Service Provider) type service that realizes the processing function only by the execution instruction and result acquisition. It is good. Note that the program in this embodiment includes information that is used for processing by an electronic computer and that conforms to the program (data that is not a direct command to the computer but has a property that defines the processing of the computer).

  In this embodiment, the present apparatus is configured by executing a predetermined program on a computer. However, at least a part of these processing contents may be realized by hardware.

  The multi-channel echo canceling apparatus according to the present invention can be used in a multi-channel communication conference system including a multi-channel reproduction system and a multi-channel sound pickup system.

10, 20, 30, 40 Multi-channel echo canceller 2 Speaker 3 Microphone 100, 110 Received signal vector conversion unit 200, 210, 220, 230 Echo replica generation unit 300 Echo replica spatial interpolation unit 400 Inverse FFT unit 500, 510 Echo cancellation Unit 600 FFT unit 800 band division voice switch unit 810 transmission / reception determination unit 820 reception side high frequency attenuation unit 830 transmission side high frequency attenuation unit 8210, 8310 high pass filter 8220, 8320 low pass filter 8230, 8330 signal attenuator 8240, 8340 signal Adder

Claims (9)

k represents time, f represents frequency, j represents frame number, i represents adaptive filter tap number, m represents speaker number, n represents microphone number, and SetN (f) is Assuming a set of microphone numbers selected from N (3 ≦ n ≦ N) microphones arranged at equal intervals in a straight line so that the spatial sampling interval becomes wider as the frequency f is lower ,
The received signal vector conversion unit converts the M channel received signal x _m (k) output from M (1 ≦ m ≦ M) speakers into the frequency domain for each channel m, and receives the received signal X _m. A received signal vector conversion step of generating (f, j);
The echo replica generation unit determines that the number of taps is I (1 ≦ i ≦ I, I is 1 or more) for the microphone n included in the SetN (f) for each frequency f from the received signal X _m (f, j). Echo replica generation step for generating spatially thinned echo replica Y ^ _{SetN (f)} (f, j) using filter coefficients H _{m, n, i} (f, j) of
The echo replica spatial interpolation unit performs spatial interpolation for each frequency f from the spatially thinned echo replica Y ^ _{SetN (f)} (f, j), and echo replica Y ^ _n (f, j). An echo replica spatial interpolation step to generate
An inverse FFT unit that converts the echo replica Y ^ _n (f, j) into the time domain to generate an echo replica y ^ _n (j);
Echo cancellation part, from the a transmission signal y _n of the N channels picked up from the N microphones that are equally spaced on a straight line _(j) echo replica y ^ n _(j), the error signal e _n An echo cancellation step to generate (k);
FFT section, the error signal _e n (k), and FFT steps are transformed into the frequency domain to generate an error signal _E n (f, j),
The echo replica generator generates a correction amount dH for the microphone n included in the SetN (f) for each frequency f from the error signal E _n (f, j) and the received signal X _m (f, j). _{m, n, i} (f, j) is obtained, and the filter coefficient H _{m, n, i} (f, j) is updated using the correction amount dH _{m, n, i} (f, j). An update step;
A multi-channel echo cancellation method comprising: k represents time, f represents frequency, j represents a frame number, i represents an adaptive filter tap number, m represents a speaker number, and SetM (f) is linearly arranged at equal intervals. Represents a set of speaker numbers selected so that the spatial sampling interval becomes wider as the frequency f is lower from M (3 ≦ m ≦ M) speakers , n represents a microphone number, and SetN (f) is Assuming a set of microphone numbers selected from N (3 ≦ n ≦ N) microphones arranged at equal intervals in a straight line so that the spatial sampling interval becomes wider as the frequency f is lower ,
Received signal vector conversion unit, a received signal of M channels output from the M speakers arranged at equal intervals on a straight line x _{m (k),} for each channel m, is converted to the frequency domain, the received signal A received signal vector conversion step of generating X _m (f, j);
The echo replica generation unit determines the number of taps for the combination of the microphone n included in the SetN (f) and the speaker m included in the SetM (f) for each frequency f from the received signal X _m (f, j). Is a spatially decimated echo replica Y ^ _{SetN (f)} (f, j) using filter coefficients H _{m, n, i} (f, j) with I (1 ≦ i ≦ I, I is 1 or more). an echo replica generation step for generating j);
The echo replica spatial interpolation unit performs spatial interpolation for each frequency f from the spatially thinned echo replica Y ^ _{SetN (f)} (f, j), and echo replica Y ^ _n (f, j). An echo replica spatial interpolation step to generate
An inverse FFT unit that converts the echo replica Y ^ _n (f, j) into the time domain to generate an echo replica y ^ _n (j);
Echo cancellation part, from the a transmission signal y _n of the N channels picked up from the N microphones that are equally spaced on a straight line _(j) echo replica y ^ n _(j), the error signal e _n An echo cancellation step to generate (k);
FFT section, the error signal _e n (k), and FFT steps are transformed into the frequency domain to generate an error signal _E n (f, j),
The echo replica generator generates the microphone n and the SetM (f) included in the SetN (f) for each frequency f from the error signal E _n (f, j) and the received signal X _m (f, j). ), The correction amount dH _{m, n, i} (f, j) is obtained, and the filter coefficient H _{m, n, i} (f, j) is obtained using the correction amount dH _{m, n, i} (f, j) _. a filter coefficient update step for updating _{n, i} (f, j);
A multi-channel echo cancellation method comprising: The multi-channel echo cancellation method according to claim 1 or 2,
S represents the microphone interval, and λ (f) represents the wavelength of frequency f.
The SetN (f) calculates the maximum natural number Q that satisfies S × Q ≦ λ (f) / 2, determines a thinning interval Q (f) that is equal to or less than the Q, and is based on the thinning interval Q (f). A multi-channel echo cancellation method characterized by being selected. The multi-channel echo cancellation method according to any one of claims 1 to 3,
S represents the microphone interval, and λ (f) represents the wavelength of frequency f.
The echo replica space interpolation step obtains a maximum natural number Q that satisfies S × Q ≦ λ (f) / 2 for each frequency f, and determines a thinning interval Q (f) that is equal to or less than Q.

A multi-channel echo cancellation method characterized by performing spatial interpolation as described above. The multi-channel echo cancellation method according to any one of claims 1 to 4,
The filter coefficient H _{m, n, i} (f, j) is variable for each frequency f, and the number of taps of the filter coefficient corresponding to a frequency exceeding a predetermined frequency corresponds to a frequency equal to or lower than the predetermined frequency. A multi-channel echo cancellation method characterized by being set to be smaller than the number of coefficient taps. A multi-channel echo cancellation method according to any of claims 1 to 5,
A band division voice switch unit receives the reception signal x _m (k) for each channel m, and a reception side high frequency attenuation step for attenuating a high frequency component;
The band division voice switch unit transmits the error signal e _n (k) for each channel n by a transmitting side high frequency attenuation step for attenuating a high frequency component;
The band division speech switch unit, using the low-frequency component of the received signal x the low-frequency component of _{m (k)} the error signal e _{n (k),} the handset determination step of performing handset judgment, further Yes And
In the transmission / reception determination step, when the reception state is determined, the reception side high frequency attenuation step is executed, and when the transmission state is determined, the transmission side high frequency attenuation step is executed. A characteristic multi-channel echo cancellation method. k represents time, f represents frequency, j represents frame number, i represents adaptive filter tap number, m represents speaker number, n represents microphone number, and SetN (f) is Assuming a set of microphone numbers selected from N (3 ≦ n ≦ N) microphones arranged at equal intervals in a straight line so that the spatial sampling interval becomes wider as the frequency f is lower ,
M (1 ≦ m ≦ M) speakers,
N microphones arranged at equal intervals on a straight line;
An M-channel received signal x _m (k) output from the speaker is converted into a frequency domain for each channel m to generate a received signal X _m (f, j);
When the received signal X _m (f, j) is input, the number of taps of the microphone n included in the SetN (f) is I (() for each frequency f from the received signal X _m (f, j). 1 ≦ i ≦ I, where I is 1 or more) using a filter coefficient H _{m, n, i} (f, j), spatially thinned echo replica Y ^ _{SetN (f)} (f, j) When the error signal E _n (f, j) is input, the SetN (f) is generated for each frequency f from the error signal E _n (f, j) and the received signal X _m (f, j). ), The correction amount dH _{m, n, i} (f, j) is obtained, and the filter coefficient H _{m, n, i} is calculated using the correction amount dH _{m, n, i} (f, j) _. an echo replica generation unit for updating _i (f, j);
Echo replica spatial interpolation for performing spatial interpolation for each frequency f from the spatially thinned echo replica Y ^ _{SetN (f)} (f, j) to generate echo replica Y ^ _n (f, j) And
An inverse FFT unit that converts the echo replica Y ^ _n (f, j) into a time domain to generate an echo replica y ^ _n (j);
From transmission signal _y n of the N channels picked up (j) and the echo replica _{y ^} n (j) from the microphone, the echo cancellation unit which generates an error signal _e n (k),
The error signal _e n (k), and the FFT unit for converting the frequency domain, the error signal _E n (f, j) to produce a,
A multi-channel echo canceling apparatus comprising: k represents time, f represents frequency, j represents a frame number, i represents an adaptive filter tap number, m represents a speaker number, and SetM (f) is linearly arranged at equal intervals. Represents a set of speaker numbers selected so that the spatial sampling interval becomes wider as the frequency f is lower from M (3 ≦ m ≦ M) speakers , n represents a microphone number, and SetN (f) is Assuming a set of microphone numbers selected from N (3 ≦ n ≦ N) microphones arranged at equal intervals in a straight line so that the spatial sampling interval becomes wider as the frequency f is lower ,
M speakers arranged at equal intervals on a straight line;
N microphones arranged at equal intervals on a straight line;
An M-channel received signal x _m (k) output from the speaker is converted into a frequency domain for each channel m to generate a received signal X _m (f, j);
When the received signal X _m (f, j) is input, the microphone n included in the SetN (f) and the SetM (f) are set for each frequency f from the received signal X _m (f, j). Echoes spatially thinned using filter coefficients H _{m, n, i} (f, j) with the number of taps I (1 ≦ i ≦ I, I is 1 or more) for combinations of speakers m included When the replica Y ^ _{SetN (f)} (f, j) is generated and the error signal E _n (f, j) is input, the error signal E _n (f, j) and the received signal X _m (f, j) j) for each frequency f, a correction amount dH _{m, n, i} (f, j) is obtained for the combination of the microphone n included in the SetN (f) and the speaker m included in the SetM (f). using the correction amount _{dH m, n,} i and (f, j), the full An echo replica generator for updating filter coefficients _{H m, n,} i and (f, j),
Echo replica spatial interpolation for performing spatial interpolation for each frequency f from the spatially thinned echo replica Y ^ _{SetN (f)} (f, j) to generate echo replica Y ^ _n (f, j) And
An inverse FFT unit that converts the echo replica Y ^ _n (f, j) into a time domain to generate an echo replica y ^ _n (j);
From transmission signal _y n of the N channels picked up (j) and the echo replica _{y ^} n (j) from the microphone, the echo cancellation unit which generates an error signal _e n (k),
The error signal _e n (k), and the FFT unit for converting the frequency domain, the error signal _E n (f, j) to produce a,
A multi-channel echo canceling apparatus comprising:   The program for making a computer perform each step of the multichannel echo cancellation method described in any one of Claim 1 to 6.

Images