JP2013255155A - Multichannel echo cancellation device and multichannel echo cancellation method and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To reduce the amount of arithmetic operation necessary to cancel an acoustic echo.SOLUTION: A frequency conversion unit 11 converts a reproduction signal to a frequency domain. A wavenumber conversion unit 12 converts the reproduction signal in frequency domain to a wavenumber domain. An echo replica generation unit 22 multiplies the reproduction signal in wavenumber domain by an adaptive filter coefficient within the effective range of wavenumbers to generate an echo replica in wavenumber domain. An inverse wavenumber conversion unit 31 converts the echo replica in wavenumber domain to a time domain. An inverse frequency conversion unit 32 converts the echo replica in frequency domain to a time domain. An echo cancellation unit 33 subtracts the echo replica in time domain from a sound pickup signal to generate an error signal. An error frequency conversion unit 41 converts the error signal to a frequency domain. An error wavenumber conversion unit 42 converts the error signal in frequency domain to a wavenumber domain. The echo replica generation unit 22 updates the adaptive filter coefficient in wavenumber domain by using a correction amount calculated within the effective range of wavenumbers from the reproduction signal in wavenumber domain and the error signal.

Description

  The present invention relates to a multi-channel echo cancellation technique for canceling acoustic echo in a multi-channel loudspeaker communication system.

  In recent years, the development of a multi-channel loudspeaker type two-way communication conference system that can provide a more natural calling environment has been progressing against the background of speeding up and capacity increase of IP communication. Multi-channel playback technology is progressing in the direction of expanding the number of channels from stereo to 5.1 channels. However, the listening area where the sound is reproduced with a high three-dimensional effect is limited, and it has become a sweet spot, and the three-dimensional sound is greatly reduced outside the sweet spot. Therefore, there is a need for multi-channel playback technology with a wide listening area that can provide all participants with a three-dimensional sound.

  In recent years, research on Wave Field Synthesis (hereinafter abbreviated as WFS) has been advanced as such a multi-channel reproduction technique (see Non-Patent Document 1). WFS requires several tens of microphones and several tens of speakers to re-synthesize a sound wave surface acquired at one point at another point.

  When applying WFS to a two-way video / audio communication conference system, in order to realize a comfortable call environment, signal components that sneak into tens to hundreds of microphones acoustically from tens to hundreds of speakers are used. It is necessary to erase from the collected sound signal. A wave number domain adaptive algorithm has been proposed as an acoustic echo canceller algorithm that efficiently performs this processing (see Non-Patent Document 2).

  The algorithm described in Non-Patent Document 2 has a filter coefficient of an adaptive filter in the wave number domain. A multi-channel communication conference system to which this algorithm is applied comprises a P-channel playback system and a P-channel sound collection system. However, P is an integer of 2 or more. The P channel reproduction signal is reproduced as an acoustic signal by a speaker, and goes around the microphone through an acoustic echo path. The reproduction signal is converted into the frequency domain, and further converted into the wave number domain for each frequency. An echo replica is generated from the reproduced signal converted into the wave number domain and the adaptive filter coefficient in the wave number domain. The echo replica is converted into the frequency domain and further converted into a P-channel time signal in the time domain. Then, the echo is eliminated by subtracting the echo replica converted to the time domain from the microphone sound pickup signal. The error signal from which the echo is eliminated from the collected sound signal is converted to the frequency domain and further converted to the wave number domain. Then, the adaptive filter coefficient is updated from the reproduction signal vector and the error signal for the P channel in the wave number domain.

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. M. Schneider, W. Kellermann, “A Wave-domain model for acoustic MIMO systems with reduced complexity”, 2011 Joint Workshop on Hands-free Speech Communication and Microphone arrays, pp.133-138, 2012.

  In an acoustic echo canceller in a multi-channel loudspeaker communication system, the number of echo paths in the adaptive filter is proportional to the square of the number of channels. Even if the wavenumber domain adaptive algorithm described in Non-Patent Document 2 is used, the amount of calculation proportional to the number of channels is required as compared with the amount of calculation of the monaural echo canceller, and the amount of calculation becomes enormous.

  The present invention has been made in view of these points, and an object thereof is to reduce the amount of calculation for eliminating acoustic echoes in a multi-channel loudspeaker communication system.

  In order to solve the above problems, the multi-channel echo canceller of the present invention includes P speakers, P microphones, a frequency converter, a wave number converter, a wave number limiter, a multiplier, and a wave number zero padding unit. A wave number conversion unit, an inverse frequency conversion unit, an echo cancellation unit, an error frequency conversion unit, an error wave number conversion unit, an error wave number limitation unit, a correction amount calculation unit, and a filter coefficient update unit are provided. The P speakers are arranged at equal intervals on a straight line. P microphones are arranged at equal intervals on a straight line. The frequency conversion unit converts the P channel reproduction signal output from the speaker into a frequency domain reproduction signal. The wave number conversion unit converts the frequency domain reproduction signal into a wave number domain reproduction signal. The wave number limiting unit calculates the effective range of the wave number for each frequency, and limits the wave number domain reproduction signal to the effective range. The multiplication unit multiplies the wave number domain effective reproduction signal within the effective range by the wave number domain adaptive filter coefficient to generate a wave number domain echo replica. The wave number zero padding unit sets an echo replica at a wave number outside the effective range to zero. The inverse wave number conversion unit converts the wave number domain echo replica into a frequency domain echo replica. The inverse frequency transform unit transforms the frequency domain echo replica into a time domain echo replica. The echo canceller subtracts the time domain echo replica from the P-channel sound pickup signal picked up from the microphone to generate an error signal. The error frequency conversion unit converts the error signal into a frequency domain error signal. The error wave number converter converts the frequency domain error signal into a wave number domain error signal. The error wavenumber limiting unit limits the wavenumber domain error signal to the effective range calculated for each frequency. The correction amount calculation unit calculates the correction amount of the wave number domain adaptive filter coefficient using the wave number domain reproduction signal and the wave number domain error signal within the effective range. The filter coefficient updating unit updates the wave number domain adaptive filter coefficient using the correction amount.

  According to the multi-channel echo cancellation technique of the present invention, it is possible to reduce the amount of calculation for canceling the acoustic echo while maintaining the echo cancellation performance in the multi-channel loudspeaker communication system.

The figure which illustrates the function structure of the conventional multichannel echo cancellation apparatus. The figure which illustrates the function structure of the conventional echo replica production | generation part. The figure which illustrates the processing flow of the conventional multichannel echo cancellation apparatus. The figure which illustrates the processing flow of the conventional echo replica production | generation part. The figure for demonstrating the synthetic | combination method of an error signal. The figure for demonstrating sampling of the plane wave of a single frequency. The figure for demonstrating the range where a signal component exists in a frequency-wavenumber area | region. The figure which illustrates the function structure of the multichannel echo cancellation apparatus which concerns on 1st Embodiment. The figure which illustrates the functional composition of the echo replica generation part concerning a 1st embodiment. The figure which illustrates the processing flow of the echo replica production | generation part which concerns on 1st Embodiment. The figure for demonstrating the relationship between generation | occurrence | production of space aliasing and a microphone space | interval. The figure which shows the result of having simulated the echo cancellation amount of the multichannel echo cancellation apparatus which concerns on 1st Embodiment. The figure which shows the result of having simulated the echo cancellation amount of the multichannel echo cancellation apparatus which concerns on 1st Embodiment. The figure which shows the calculation amount reduction effect of the multichannel echo cancellation apparatus which concerns on 1st Embodiment. The figure which illustrates the function structure of the multichannel echo cancellation apparatus which concerns on 2nd Embodiment. The figure which illustrates the function structure of the echo replica production | generation part which concerns on 2nd Embodiment. The figure which illustrates the processing flow of the echo replica production | generation part which concerns on 2nd Embodiment. The figure which shows the result of having simulated the echo cancellation amount of the multichannel echo cancellation apparatus which concerns on 2nd Embodiment. The figure which shows the result of having simulated the echo cancellation amount of the multichannel echo cancellation apparatus which concerns on 2nd Embodiment. The figure which shows the calculation amount reduction effect of the multichannel echo cancellation apparatus which concerns on 2nd Embodiment.

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

を表すものとする。また、A^は、

を表すものとする。 In this description, _A is

. A ^ is

.

[Conventional multi-channel echo canceller]
<Overview>
Prior to the description of the embodiment, the processing of the conventional multi-channel echo canceller will be described in detail. This multi-channel echo canceller cancels acoustic echoes from the collected sound signal by a wavenumber domain adaptive algorithm described in Non-Patent Document 2.

<Configuration>
FIG. 1 shows a configuration example of a conventional multi-channel echo canceller 9. Conventional multi-channel echo canceller 9, P (≧ 2) pieces of input terminals ₁ 1, ..., 1 _P and P number of the speaker 2 _1, ..., 2 _P and P number of microphones 3 _1, ..., 3 _P P outputs terminals 4 _1, ..., 4 _P and P number of the frequency converting unit 11 _1, ..., 11 _P and the wave number converter 12 and the P number of echo replica generation unit 21 _1, ..., 21 _P opposite wavenumber converter 31 and the P number of inverse frequency transform unit 32 _1, ..., 32 _P and the echo canceling portion 33 and the P number of error frequency converter 41 _1, ..., 41 _P and the error-wavenumber conversion portion 42 and the frame synthesizing portion 91 Is provided. P microphones 3 ₁ ,..., 3 _P are arranged at equal intervals on a straight line.

FIG. 2 shows a configuration example of the echo replica generation unit 21 _p (1 ≦ p ≦ P) provided in the multi-channel echo canceller 9. The echo replica generation unit 21 _p includes a correction amount calculation unit 211, a filter coefficient update unit 213, and a multiplication unit 215.

<Operation>
With reference to FIGS. 3 and 4, an example of the operation of the multi-channel echo canceller 9 will be described in detail according to the order of procedures actually performed. In the following description, each signal is blocked by 1 frame 2F samples and shift amount F / D samples. However, F is a natural number and D is a natural number that divides F. In order to simplify and speed up the FFT calculation, F is generally a power of 2. In this example, it is assumed that D ≧ 2. Also, n is the time and i is the frame number.

The frequency converter 11 _p (1 ≦ p ≦ P) blocks one frame (2F samples) of the P channel reproduction signal x (p, n) into blocks according to the equation (1), and reproduces the reproduction signal vector x (p, p, p). i).

  Subsequently, the reproduction signal vector x (p, i) is converted into the frequency domain according to the equation (2) to generate the reproduction signal vector X (p, i) in the frequency domain (S11).

Here, f is a frequency index. Assuming that the sampling frequency of the reproduction signal is f _s , X _f (p, i) in equation (2) is a component of frequency f _s f / 2F [Hz].

The wave number conversion unit 12 converts the reproduction signal X _f (p, i) in the frequency domain into a reproduction signal _X _f (k, i) in the wave number domain for each frequency index f (S12). When the number of channels P of the reproduction signal is an even number, it is converted into the wave number domain according to the equation (3). Here, P = 2K.

  On the other hand, when the number of channels P of the reproduction signal is an odd number, it is converted into the wave number domain by Expression (4). Here, P = 2K + 1.

  The conversion from the frequency domain to the wave number domain is usually performed at high speed by a fast Fourier transform having a power of two. In the following description, a case where the number of channels P is an even number will be described as an example.

The multiplication unit 215 included in the echo replica generation unit 21 _p (1 ≦ p ≦ P) has the full wavenumber index, ie, −K ≦ k <K, for the frequency index f satisfying f ≦ F, the reproduction signal _X of the wavenumber domain Multiply _f (k, i) by the adaptive filter coefficient _H _f (k, k, i) to generate an echo replica _Y ^ _f (k, i) in the wave number domain (S215). This complex multiplication can be expressed as Equation (5).

  At this time, adjacent spatial frequency components may be included as in Expression (6). In Non-Patent Document 2, 1 or 2 is recommended as δ.

The inverse wave number transforming unit 31 transforms the echo replica_Y ^ _f (k, i) in the wave number domain into the frequency domain according to the equation (7) to generate the frequency domain echo replica Y ^ _f (p, i) ( S31).

For the frequency index f where f> F, a frequency domain echo replica Y ^ _f (p, i) is generated according to equation (8) using the symmetry of the Fourier transform result of the real signal. Here, conj (·) means taking a complex conjugate of •.

The inverse frequency transform unit 32 _p (1 ≦ p ≦ P) performs an inverse fast Fourier transform on the frequency domain echo replica Y ^ _f (p, i) according to the equation (9), and the time domain echo replica signal vector y ^ Conversion into (p, i) is performed (S32). Here, 0 _F is an F × F zero matrix, and _IF is an F × F unit matrix.

  The echo canceller 33 blocks one frame (2F samples) of the P channel sound pickup signal y (p, n) according to the equation (10) to generate a sound pickup signal vector y (p, i).

  Subsequently, according to the equation (11), the echo replica signal vector y ^ (p, i) is subtracted from the collected sound signal vector y (p, i) to generate an error signal vector e (p, i) (S33). .

In Equation (1) and Equation (10), when D ≧ 2 when each signal is framed, the frame synthesis unit 91 obtains the error signal obtained by frame number i and the previous frame number. It is necessary to synthesize the error signals through a windowing process (S91). The windowing process will be described with reference to FIG. When D = 2, the combined error signal vector e ′ (p, i) is calculated according to the equation (12) using the window W _H having the length F / 2. Here, 0 _{F / 2} is a zero matrix of (F / 2) × (F / 2), and I _{F / 2} is a unit matrix of (F / 2) × (F / 2). The window W _H is, for example, a Hanning window.

The error frequency conversion unit 41 _p (1 ≦ p ≦ P) converts the error signal vector e (p, i) into the frequency domain according to the equation (13), and the frequency domain error signal vector E (p, i). Is generated (S41).

The error wave number converter 42 converts the error signal E _f (p, i) in the frequency domain into the wave number domain according to the equation (14) to generate the error signal _E _f (k, i) in the wave number domain ( S42).

The correction amount calculation unit 211 included in the echo replica generation unit 21 _p (1 ≦ p ≦ P) has the frequency index f satisfying f ≦ F according to the equation (15) in the full wave number index (−K ≦ k <K). The correction amount d_H _f (k, k + dk, i) of the adaptive filter coefficient is calculated (S211). However, dk is −δ ≦ df ≦ δ.

However, conj (•) means taking a complex conjugate of •. ρ is a minute positive constant for preventing the denominator from becoming zero. _Z _f (k, i) is the total power of the reproduction signal _X (k, i) whose wave number index for each frequency component is (k-δ) to (k + δ), and the correction amount d_H (k, k + dk, i) is corrected. _Z _f (k, i) is obtained by Expression (16). However, β is a smoothing constant for taking a short time average in power calculation, and takes a value from 0 to 1.

The filter coefficient update unit 213 included in the echo replica generation unit 21 _p (1 ≦ p ≦ P) uses the correction amount d_H _f (k, k + dk, i) according to the equation (17) to apply the adaptive filter coefficient _H _f ( k, k, i) is updated (S213). However, μ is a step size and takes a value from 0 to 1.

  Thus, in the conventional multi-channel echo canceller, for example, when the number P of speakers and microphones is 100 and δ = 0 is set, the number of echo paths in the adaptive filter is 100 channels, that is, a monaural echo canceller. It becomes 100 times. Furthermore, when δ = 2 is set, the amount reaches 100 × (1 + 2 × δ (= 2)) = 500 channels, and the amount of calculation becomes enormous. An object of the present invention is to reduce the amount of calculation for eliminating acoustic echo in a multi-channel loudspeaker communication system having a large number of echo paths.

[Points of Invention]
The operation principle of the present invention will be described. First, consider how a single-frequency plane wave is represented in the wave number domain. When viewed in the frequency-wavenumber space, it is known that the existence range of waves is limited depending on the frequency (for details, see “T. Ajdler, L. Sbaiz, and M. Vetterli,“ Dynamic measurement of room impulse responses ”. using a moving microphone ", J. Acoust. Soc. Am., 122 (3), 2007, 1636-1645. (reference 1)"). With reference to FIG. 6, a plane wave having a single frequency f ₀ [Hz] incident at an angle α is considered. In this example, microphone elements are arranged on a straight line at equal intervals d. When the microphone element array is taken on the u-axis, the time variation of the sound pressure on the u-axis can be expressed by Expression (18), where t is the time.

However, the angular frequency ω ₀ and the wave number φ ₀ have the relationship of Equation (19), where the velocity of sound is veloc.

  When the sound pressure on the u-t axis is converted into the frequency-wave number domain, it can be expressed by equation (20).

  According to Expression (20), it can be seen that a plane wave having a single frequency in the time-space domain is represented by one point in the frequency-wave number domain.

  A plane wave having the same frequency component at all frequencies and represented by Expression (21) in the time-space domain is represented by Expression (22) in the frequency-wave number domain and is represented by a straight line.

  Since the incident angle α of the plane wave is in the range of 0 to 180 degrees, the existence range of the plane wave is limited to a range satisfying the expression (23) in the frequency-wave number region.

Here, f ′ is a frequency [Hz] corresponding to the angular frequency ω.

  FIG. 7 illustrates the existence range of waves in the frequency-wave number region. The vertical axis is the wave number φ, and the horizontal axis is the angular frequency ω. In this case, the wave exists only in a range surrounded by a straight line having a slope of 1 / veloc and a straight line having a slope of -1 / veloc.

Sampling by the actual microphone array is discrete. For the time direction, use the 2F point-FFT with a sampling frequency of f _s and a frame length of 2F, and for the spatial direction, use the 2K point-FFT with a microphone interval of d and the number of microphones of 2K. Is in the range of 0 to f _s / 2 [Hz], and the range of the wave number φ is from −π / d to π / d.

  In the wavenumber domain adaptive algorithm described in Non-Patent Document 2, in the process of obtaining an echo replica in the wavenumber domain and the process of updating the adaptive filter coefficient, −K ≦ corresponding to the range −π / d to π / d of the total wavenumber φ. Processing is performed in the range of k <K. However, according to the above knowledge, the range in which the signal component exists in the frequency-wavenumber region becomes narrower as the frequency of the sound wave is lower. In the present invention, based on the relationship between the frequency and the wave number range, the echo replica calculation process and the adaptive filter coefficient update process are performed only in the range where the signal component exists. The amount of calculation can be reduced by narrowing down these processes to the signal component existence range.

[First Embodiment]
<Configuration>
FIG. 8 shows a configuration example of the multi-channel echo cancellation apparatus 10 of the first embodiment. Multi-channel echo canceller 10, P (≧ 2) pieces of input terminals ₁ 1, ..., 1 _P and P number of the speaker 2 _1, ..., 2 _P and P number of microphones 3 _1, ..., 3 _P and P number of output terminals 4 _1, ..., 4 _P and P number of the frequency converting unit 11 _1, ..., 11 _P and the wave number converter 12 and the P number of echo replica generation unit 22 _1, ..., 22 _P and the inverse-wavenumber conversion portion 31 and P number of the inverse frequency transformation unit 32 _1, ..., 32 _P and the echo canceling portion 33 and the P number of error frequency converter 41 _1, ..., and a 41 _P and error-wavenumber conversion portion 42 and the frame synthesizing portion 91 . P speakers 2 ₁ ,..., 2 _P are arranged at equal intervals on a straight line. P microphones 3 ₁ ,..., 3 _P are arranged at equal intervals on a straight line.

  The frequency conversion unit 11, the inverse frequency conversion unit 32, and the error frequency conversion unit 41 perform frequency conversion or inverse frequency conversion in the same manner as the conventional multichannel echo canceller 9. Although the example using the FFT process or the inverse FFT process has been described above, the frequency conversion or the inverse frequency conversion can be performed by a process other than the FFT.

FIG. 9 shows a configuration example of the echo replica generation unit 22 _p (1 ≦ p ≦ P) included in the multi-channel echo cancellation apparatus 10. The echo replica generation unit 22 _p includes a correction amount calculation unit 211, a filter coefficient update unit 213, a multiplication unit 215, a wave number limiting unit 216, an error wave number limiting unit 217, and a wave number zero padding unit 218.

The only difference between the conventional multi-channel echo canceller 9 and the multi-channel echo canceller 10 of the first embodiment is that the configuration of the echo replica generator is different. The difference between the echo replica generation unit 21 _{p included in the} multichannel echo cancellation apparatus 9 and the echo replica generation unit 22 _{p included in the} multichannel echo cancellation apparatus 10 is that the wave number limiting unit 216, the error wave number limiting unit 217, and the wave number 0 filling unit It is a point provided with 218.

<Operation>
With reference to FIG. 10, an example of the operation of the echo replica generation unit 22 _p will be described in detail according to the order of procedures actually performed.

The wave number limiting unit 216 included in the echo replica generation unit 22 _p calculates an effective wave number index max_k (f) for each frequency index f. Then, the reproduction signal _X _f (k, i) in the wavenumber domain is limited to the range of −max_k (f) ≦ k ≦ max_k (f), and the effective reproduction signal _X _f (k, i) in the wavenumber domain is Generate (S216).

The wave with the frequency index f corresponds to the wave with the frequency f _s f / 2F [Hz]. Since the angular frequency ω of this wave can be expressed by equation (24), the upper limit of the wave number of this wave can be expressed by equation (25).

  Since the wave number index K corresponds to π / d, the upper limit max_k (f) of the corresponding wave number index can be expressed by Expression (26). However, ceil (•) is a function that rounds up • to an integer.

  As described above, the effective wave number index range for the frequency index f is −max_k (f) ≦ k ≦ max_k (f).

The multiplier 215 included in the echo replica generator 22 _p has an effective reproduction signal _X _f (k, i in the wavenumber domain in the range of −max_k (f) ≦ k ≦ max_k (f) according to the above equation (5). ) Is multiplied by an adaptive filter coefficient _H _f (k, k, i) to generate an echo replica _Y ^ _f (k, i) in the wave number domain (S215). When including adjacent spatial frequency components, the effective reproduction signal _X _f (k, i) in the wavenumber domain is in the range of −max_k (f) ≦ k ≦ max_k (f) according to the above equation (6). By multiplying the adaptive filter coefficient _H _f (k, k, i), the wave number domain echo replica _Y ^ _f (k, i) may be generated.

The wave number zero padding unit 218 included in the echo replica generation unit 22 _p is _Y for the range of k <−max_k (f) and k> max_k (f) for the echo replica_Y ^ _f (k, i) in the wave number domain. ^ _f (k, i) = 0 is set (S218).

The error wave number limiting unit 217 included in the echo replica generation unit 22 _p sets the error signal _E _f (k, i) in the wave number domain to a range of −max_k (f) ≦ k ≦ max_k (f) for each frequency index f. In a limited manner, an effective error signal _E _f (k, i) in the wave number domain is generated (S217). As the effective wave number index max_k (f), the value obtained by the wave number limiting unit 216 can be used.

The correction amount calculation unit 211 included in the echo replica generation unit 22 _p has the above-described formula (with respect to the frequency index f satisfying f ≦ F) within the range where the wave number index k is −max_k (f) ≦ k ≦ max_k (f) ( According to 15), the correction amount d_H _f (k, k + dk, i) (−δ ≦ dk ≦ δ) is calculated (S211).

The filter coefficient update unit 213 included in the echo replica generation unit 22 _p uses the adaptive filter coefficient correction amount d_H _f (k, k + dk, i) according to the above formulas (16) and (17). H _f (k, k, i) is updated (S213).

The relationship between the occurrence of spatial aliasing and the microphone interval will be described with reference to FIG. When the sampling frequency is f _s , the wave number is πf _s / veloc for a wave having a frequency f _s / 2 [Hz] that is the upper limit of time sampling. Spatial aliasing does not occur when the microphone spacing d is sufficiently small and π / d is greater than this value. However, spatial aliasing occurs when the microphone interval d is relatively long and π / d is smaller than this value. The microphone interval d and the sampling frequency f _s must be set to values that do not cause spatial aliasing.

<Simulation results>
In order to confirm the effect of the first embodiment, an echo cancellation amount was simulated. In a room with a reverberation time of 250 milliseconds, a speaker array in which 32 speaker elements are arranged in a straight line with a spacing of 2 cm and a microphone array in which 32 microphone elements are arranged in a straight line with a spacing of 2 cm are parallel by 40 cm. Arranged. The sampling frequency was set to 16 kHz, and the frame length was set to 2F = 2048. The reproduction signal was generated by separately simulating the sound picked up by 32 microphones using pink noise as the sound source. In this environment, impulse responses were measured for all echo paths between the speaker and microphone.

  12 and 13 show the reverberation echo cancellation results. FIG. 12 (A1) is a graph plotting the collected sound signal level and the signal level after the echo cancellation process when the echo cancellation process is performed using the conventional multi-channel echo cancellation method for the first channel among the 32 channels. It is. The dotted line is the collected sound signal level, and the solid line is the signal level after the echo cancellation processing. FIG. 12 (A2) is a diagram in which the collected sound signal level and the signal level after echo cancellation processing are plotted for the first channel when echo cancellation processing is performed using the multi-channel echo cancellation method of the first embodiment. is there. FIG. 12B1 shows the result of canceling echo for the third channel using the conventional method, and FIG. 12B2 shows the result of canceling echo for the third channel using the method of the first embodiment. . FIG. 12C1 shows the result of canceling echo for the fifth channel using the conventional method, and FIG. 12C2 shows the result of canceling echo for the fifth channel using the method of the first embodiment. . FIG. 13D1 shows the result of canceling echo for the seventh channel using the conventional method, and FIG. 13D2 shows the result of canceling echo for the seventh channel using the method of the first embodiment. . FIG. 13E1 shows the result of canceling echo for the ninth channel using the conventional method, and FIG. 13E2 shows the result of canceling echo for the ninth channel using the method of the first embodiment. .

  The object of the present invention is to reduce the amount of computation of echo cancellation while maintaining the echo cancellation performance. Therefore, in any channel, the amount of echo cancellation by the conventional method and the amount of echo cancellation by the method of the first embodiment are reduced. It is desirable that the signal levels after the echo cancellation processing plotted with a solid line are equivalent power [dB].

  In the third, fifth, seventh and ninth channels, it can be seen that the method of the first embodiment can cancel the echo in the same manner as the conventional method. In the first channel, the echo cancellation amount by the method of the first embodiment is deteriorated more than the echo cancellation amount by the conventional method, but the deterioration width is only 2 dB.

  FIG. 14 shows the calculation amount reduction effect by the multi-channel echo cancellation method of the first embodiment. The calculation amount is a complex multiplication count necessary for signal processing for one frame. The calculation amount of each processing unit according to the conventional method is as follows.

Frequency converter 11 (S11): 2F point-FFT × P channel Wave number converter 12 (S12): P point-FFT × F band Echo replica generator 21 (S21A): P channel × (1 + 2δ) × F band × C block / inverse wave number conversion unit 31 (S31): P point-inverse FFT × F band / inverse frequency conversion unit 32 (S32): 2P point-inverse FFT × P channel / error frequency conversion unit 41 (S41) ): 2F point−FFT × P channel / error wave number converting unit 42 (S42): P point−FFT × F band echo replica generating unit 21 (S21B): {P channel × (1 + 2δ) +2} × F According to the multi-channel echo canceller of the band × C block first embodiment, the calculation amount of the echo replica generation unit 21 (S21A, S21B) is halved.

  FIG. 14 is a diagram illustrating the calculation amount reduction effect by the method of the first embodiment. When the vertical axis is the complex number of multiplications and the horizontal axis is the number of channels, the number of complex multiplications necessary for signal processing for one frame is the number of microphones P = 32, 64, 128, 256, δ = 1, and the frame length is 2F = 2048. The calculation amount by the conventional method is plotted by a dotted line, and the calculation amount by the method of the first embodiment is plotted by a solid line. The multiplication amount of 2F point-FFT was calculated as Flog2 (2F). It can be seen that for any number of channels, the amount of computation by the method of the first embodiment can be reduced by more than 15% than the amount of computation by the conventional method.

<Effect>
As described above, the multi-channel echo canceller of the present invention focuses on the fact that the effective wave number region is limited as the frequency is lower in the frequency-wave number region. Is limited to an effective wavenumber region. As a result, it is possible to reduce the overall calculation amount in a multi-channel loudspeaker communication system composed of a large number of speakers and microphones while maintaining the echo cancellation performance.

[Second Embodiment]
<Overview>
The second embodiment of the present invention is a case where the concept of the frequency domain adaptive algorithm is combined with the adaptive filter in the wave number domain. For details on frequency domain adaptation algorithms, see “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. (Reference 2). Reference 2 shows a method of dividing the adaptive filter in the time direction. By this division, the adaptive filter length can be extended while suppressing the processing delay to match the length of the impulse response to be processed, so that the echo canceller can be adapted to a room with longer reverberation.

<Configuration>
FIG. 15 shows a configuration example of the multi-channel echo canceling apparatus 10 ′ of the second embodiment. The multi-channel echo canceling apparatus 10 ′ is similar to the multi-channel echo canceling apparatus 10 of the first embodiment in that P (≧ 2) input terminals 1 ₁ ,..., 1 _P and P speakers 2 ₁ ,. 2 _P and P microphones 3 ₁ , ..., 3 _P and P output terminals 4 ₁ , ..., 4 _P and P frequency converters 11 ₁ , ..., 11 _P , wave number converter 12 and inverse wave number converter parts 31 and P number of the inverse frequency transformation unit 32 _1, ..., 32 _P and the echo canceling portion 33 and the P number of error frequency converter 41 _1, ..., 41 _P and the error-wavenumber conversion portion 42 and the frame synthesizing portion 91 Prepare. P number of echo replica generation unit 22 _1, ..., provided in place of 22 _P, P number of echo replica generation unit 23 _1, ..., a 23 _P.

FIG. 16 shows a configuration example of the echo replica generation unit 23 _p (1 ≦ p ≦ P) included in the multichannel echo canceller 10 ′. Similar to the echo replica generation unit 22 _{p of} the first embodiment, the echo replica generation unit 23 _p includes a wave number limiting unit 216, an error wave number limiting unit 217, and a wave number zero padding unit 218. A correction amount calculation unit 221 is provided instead of the correction amount calculation unit 211, a filter coefficient update unit 223 is provided instead of the filter coefficient update unit 213, and a multiplication unit 225 is provided instead of the multiplication unit 215. Furthermore, a reception storage unit 219 is provided.

The only difference between the multi-channel echo cancellation apparatus 10 of the first embodiment and the multi-channel echo cancellation apparatus 10 'of the second embodiment is that the configuration of the echo replica generation unit is different. The difference between the echo replica generation unit 22 _{p included} in the multichannel echo cancellation apparatus 10 and the echo replica generation unit 23 _{p included in the} multichannel echo cancellation apparatus 10 ′ is that it further includes a reception accumulation unit 219, and a correction amount calculation unit The processing of the filter coefficient updating unit and the multiplication unit is different.

<Operation>
With reference to FIG. 17, an operation example of the echo replica generation unit 23 _p will be described in detail according to the order of procedures actually performed. In this embodiment, the adaptive filter coefficient is divided into C pieces in the time direction as _H _f (k, k, i, c) (0 ≦ c ≦ C−1).

The receiver storage section 219 included in the echo replica generator 23 _p, an effective reproduced signal _X _f (p, i) of the frequency domain to produce a wavenumber limiting portion 216 of the most recent 2C-1 or a is _X _f (p, i) to _X _f (p, i-2 (C-1)) are accumulated (S219).

The multiplication unit 225 included in the echo replica generation unit 23 _p has an adaptive filter coefficient _H _f (divided in the time direction in the range of −max_k (f) ≦ k ≦ max_k (f) according to the following equation (27). k, k, i, c) (equivalent to wave number domain echo replica_Y ^ _f (k, i) Is generated (S225). Here, the frame shift amount is set to F / 2 (D = 2).

When including adjacent spatial frequency components, the adaptive filter coefficient _H _f (k, k, f) divided in the time direction in the range of −max_k (f) ≦ k ≦ max_k (f) according to the following equation (28): k, i, c) (0≤c≤C-1) is multiplied by the effective reproduction signal in the wavenumber domain and summed to generate a wavenumber domain echo replica_Y ^ _f (k, i). That's fine.

The correction amount calculation unit 221 included in the echo replica generation unit 23 _p has the following expression (in the range where the wave number index k is −max_k (f) ≦ k ≦ max_k (f) for the frequency index f where f ≦ F: According to 29), the correction amount d_H _f (k, k + dk, i, c) of the adaptive filter is calculated (S221).

However, conj (•) means taking a complex conjugate of •. ρ is a minute positive constant for preventing the denominator from becoming zero. _Z _f (k, i) is the total power of the reproduction signal _X (k, i) whose wave number index for each frequency component is (k-δ) to (k + δ), and the correction amount d_H (k, k + dk, i) is corrected. _Z _f (k, i) is obtained by Expression (30). However, β is a smoothing constant for taking a short-time average in power calculation, and takes a value from 0 to 1.

The filter coefficient updating unit 223 included in the echo replica generation unit 23 _p (1 ≦ p ≦ P) uses the correction amount d_H _f (k, k + dk, i, c) in accordance with the following equation (31). The coefficient _H _f (k, k, i, c) is updated (S223). However, μ is a step size and takes a value from 0 to 1.

<Simulation results>
In order to confirm the effect of the second embodiment, an echo cancellation amount was simulated. In a room with a reverberation time of 250 milliseconds, a speaker array in which 32 speaker elements are arranged in a straight line with a spacing of 2 cm and a microphone array in which 32 microphone elements are arranged in a straight line with a spacing of 2 cm are parallel by 40 cm. Arranged. The sampling frequency was set to 16 kHz, the frame length was set to 2F = 512, and the adaptive filter division number was set to C = 4. The reproduction signal was generated by separately simulating the sound picked up by 32 microphones using pink noise as the sound source. In this environment, impulse responses were measured for all echo paths between the speaker and microphone.

  18 and 19 show the reverberation echo cancellation results. FIG. 18A1 is a diagram in which the collected sound signal level and the signal level after echo cancellation processing are plotted for the first channel of 32 channels when echo cancellation processing is performed using the conventional multi-channel echo cancellation method. It is. The dotted line is the collected sound signal level, and the solid line is the signal level after the echo cancellation processing. FIG. 18A2 is a diagram in which the collected sound signal level and the signal level after the echo cancellation process are plotted for the first channel when the echo cancellation process is performed using the multi-channel echo cancellation method of the second embodiment. is there. FIG. 18B1 shows the result of canceling echo for the third channel using the conventional method, and FIG. 18B2 shows the result of canceling echo for the third channel using the method of the second embodiment. . FIG. 18C1 shows the result of canceling echo for the fifth channel using the conventional method, and FIG. 18C2 shows the result of canceling echo for the fifth channel using the method of the second embodiment. . FIG. 19D1 shows the result of canceling echo for the seventh channel using the conventional method, and FIG. 19D2 shows the result of canceling echo for the seventh channel using the method of the second embodiment. . FIG. 19E1 shows the result of canceling echo for the ninth channel using the conventional method, and FIG. 19E2 shows the result of canceling echo for the ninth channel using the method of the second embodiment. .

  Similar to the first embodiment, the echo cancellation amount by the conventional method and the echo cancellation amount by the method of the second embodiment are equal in any channel, that is, the signal level after the echo cancellation process plotted by the solid line However, it is desirable that the power [dB] is the same between the conventional method and the method of the second embodiment in the same channel.

  In the third, fifth, seventh and ninth channels, it can be seen that the method of the second embodiment can cancel the echo in the same manner as the conventional method. In the first channel, the echo cancellation amount by the method of the second embodiment is deteriorated more than the echo cancellation amount by the conventional method, but the deterioration width is only 2 dB. That is, it can be seen that the method of the second embodiment has the same echo cancellation performance as that of the method of the first embodiment.

  FIG. 20 is a diagram illustrating the calculation amount reduction effect by the method of the second embodiment. The number of complex multiplications required for signal processing for one frame, the number of microphones P = 32, 64, 128, 256, δ = 1, frame length 2F = 512, number of divisions, where the vertical axis is the number of complex multiplications and the horizontal axis is the number of channels In the case of C = 4, the calculation amount by the conventional method is plotted by a dotted line, and the calculation amount by the method of the second embodiment is plotted by a solid line. It can be seen that, regardless of the number of channels, the amount of computation by the method of the second embodiment can be reduced by 35% compared to the amount of computation by the conventional method.

<Effect>
Thus, according to the present invention, even when the wavenumber domain adaptive filter is divided in the time direction as in the multichannel echo canceller of the second embodiment, the same as the multichannel echo canceller of the first embodiment. In addition, it is possible to reduce the overall calculation amount in a multi-channel loudspeaker communication system composed of a large number of speakers and microphones while maintaining the echo cancellation performance.

[Program, recording medium]
The present invention is not limited to the above-described embodiment, and it goes without saying that modifications can be made as appropriate without departing from the spirit of the present invention. The various processes described in the above-described embodiments are not only executed in time series according to the order described, but may be executed in parallel or individually as required by the processing capability of the apparatus that executes the processes.

  When various processing functions in each device described in the above embodiment are realized by a computer, the processing contents of the functions that each device should have are described by a program. Then, by executing this program on a computer, various processing functions in each of the above devices are realized 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.

9, 10, 10 ′ Multi-channel echo canceling device 1 Input terminal 2 Speaker 3 Microphone 4 Output terminal 11 Frequency conversion unit 12 Wave number conversion unit 21, 22, 23 Echo replica generation unit 31 Inverse wave number conversion unit 32 Inverse frequency conversion unit 33 Echo Erasing unit 41 Error frequency converting unit 42 Error wave number converting unit 91 Frame synthesizing unit 211, 221 Correction amount calculating unit 213, 223 Filter coefficient updating unit 215, 225 Multiplying unit 216 Wave number limiting unit 217 Error wave number limiting unit 218 Wave number zero padding unit 219 Earphone storage unit

Claims (6)

P (P ≧ 2) speakers arranged at equal intervals on a straight line,
P microphones arranged at equal intervals on a straight line,
A frequency conversion unit for converting a reproduction signal of the P channel output from the speaker into a frequency domain reproduction signal;
A wave number converter for converting the frequency domain reproduction signal into a wave number domain reproduction signal;
Calculating a wave number effective range for each frequency, and limiting the wave number domain reproduction signal to the effective range; and
A multiplier that multiplies the wavenumber domain reproduction signal within the effective range by a wavenumber domain adaptive filter coefficient to generate a wavenumber domain echo replica;
A wave number zero padding part that sets an echo replica at a wave number outside the effective range to zero;
An inverse wavenumber converter for converting the wavenumber domain echo replica into a frequency domain echo replica;
An inverse frequency transform unit for transforming the frequency domain echo replica into a time domain echo replica;
An echo canceler that subtracts the time-domain echo replica from the collected sound signal of the P channel collected from the microphone, and generates an error signal;
An error frequency converter for converting the error signal into a frequency domain error signal;
An error wavenumber converter for converting the frequency domain error signal into a wavenumber domain error signal;
An error wavenumber limiting unit that limits the wavenumber domain error signal to the effective range calculated for each frequency;
A correction amount calculation unit for calculating a correction amount of the wave number domain adaptive filter coefficient using the wave number domain reproduction signal and the wave number domain error signal within the effective range;
A filter coefficient updating unit that updates the wavenumber domain adaptive filter coefficient using the correction amount;
A multi-channel echo canceller comprising: The multi-channel echo canceller according to claim 1,
A reception storage unit that divides the reproduction signal at predetermined time intervals and converts the reproduction signal to a wave number domain through a frequency domain, and accumulates a plurality of past wave number domain reproduction signals within the effective range;
The multi-channel echo cancellation apparatus, wherein the multiplication unit multiplies an adaptive filter coefficient corresponding to the accumulated wave number domain reproduction signal, and generates the wave number domain echo replica by taking the sum. The multi-channel echo canceller according to claim 1 or 2,
A multi-channel echo canceller that calculates the upper and lower limits of the effective range of wave numbers for each frequency as a linear function of frequency. The multi-channel echo canceller according to claim 3,
The effective range of wave numbers when the frequency is f [Hz]

A multi-channel echo canceller characterized by: A frequency conversion step of converting a reproduction signal of P channel output from P (P ≧ 2) speakers arranged at equal intervals on a straight line into a frequency domain reproduction signal;
A wave number conversion step for converting the frequency domain reproduction signal into a wave number domain reproduction signal;
Calculating a wave number effective range for each frequency, and limiting the wave number domain reproduction signal to the effective range;
A multiplication step of multiplying the wavenumber domain reproduction signal within the effective range by a wavenumber domain adaptive filter coefficient to generate a wavenumber domain echo replica;
A wave number zero filling step in which an echo replica at a wave number outside the effective range is zero;
An inverse wavenumber conversion step for converting the wavenumber domain echo replica into a frequency domain echo replica;
An inverse frequency transform step for transforming the frequency domain echo replica into a time domain echo replica;
An echo cancellation step of subtracting the time domain echo replica from the P channel sound pickup signal picked up from P microphones arranged at equal intervals on a straight line to generate an error signal;
An error frequency conversion step for converting the error signal into a frequency domain error signal;
An error wavenumber converting step of converting the frequency domain error signal into a wavenumber domain error signal;
An error wavenumber limiting step for limiting the wavenumber domain error signal to the effective range calculated for each frequency;
A correction amount calculating step of calculating a correction amount of the wave number domain adaptive filter coefficient using the wave number domain reproduction signal and the wave number domain error signal within the effective range;
A filter coefficient update step of updating the wavenumber domain adaptive filter coefficient using the correction amount;
A multi-channel echo cancellation method including:   A program for causing a computer to function as the multi-channel echo canceling device according to claim 1.

Images