JP6075783B2 - Echo canceling apparatus, echo canceling method and program - Google Patents

Description

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

  In recent years, development of a multi-channel loudspeaker type two-way communication conferencing system that can provide a more natural calling environment has progressed against the background of higher speed and higher capacity of IP communication. Multi-channel playback technology is also progressing in the direction of expanding the number of channels from stereo playback to 5.1 channel playback. 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 outside it, the three-dimensional effect of the sound is greatly reduced.

  Therefore, research on Wave Field Synthesis (hereinafter abbreviated as “WFS”) has recently been advanced as a multi-channel playback technique with a wide listening area (see Non-Patent Document 1). WFS is a technique for acquiring a sound wave surface at a certain point and recombining it at another point.

  When a WFS is applied to a two-way video / audio communication conference, in order to realize a comfortable call environment, a signal component (hereinafter referred to as “sound sneaking” from tens to hundreds of speakers to tens to hundreds of microphones). (Also called “echo”) must be erased from the microphone's collected 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). This wave number domain adaptive algorithm is an algorithm having filter coefficients of an adaptive filter in the wave number domain.

  However, as described in the explanation of the simulation result of Non-Patent Document 2, when the radiation direction of the wavefront reproduced from the speaker array is changed, the echo cancellation amount rapidly deteriorates. This situation corresponds to a case in which a speaker is switched at a remote place in two-way communication, and the radiation direction of the speaker reproduced voice after the substitution is different from the radiation direction before the substitution. The reason why the echo cancellation amount deteriorates is that when the radiation direction of the reproduction wavefront changes, adaptive filter coefficients having different wave numbers are required for echo cancellation, but the adaptive filter coefficients are almost unlearned.

  In order to realize a comfortable voice call, it is necessary to quickly reduce the residual echo regardless of the conversation state in a state where the echo path estimation and cancellation by the adaptive filter is not sufficient. Especially in the double talk state, it is necessary to reduce the residual echo without affecting the quality of transmission.

  As such a method, Non-Patent Document 3 proposes a method of estimating and canceling a residual echo included in an error signal in the wave number domain.

J. Berkhout, D de Vries, and P. Vogel, "Acoustic Control by wave field synthesis", Journal of Acoustic Society of America, 1993, vol.93, no.5, p.2764-2778 M. Schneider, W. Kellermann, "A Wave-domain model for acoustic MIMO systems with reduced complexity", 2012, 2011 Joint Workshop on Hands-free Speech Communication and Microphone arrays, pp.133-138 S. Emura et.al., "Posterior residual echo cancellation and its complexity reduction in the wave domain, Acoustic Signal Enhancement", Proceedings of IWAENC 2012, 2012, International Workshop on.

  However, in order to increase the playback volume in order to expand the listening area and increase the microphone gain in order to expand the sound collection area, it is necessary to further improve the performance of residual echo cancellation.

  Residual echoes include those based on direct waves that do not depend on reflections, and those based on reflected waves other than direct waves (diffuse residual echoes). Since the method of Non-Patent Document 3 is a model used as a base, only a residual echo due to a direct wave is targeted.

  An object of the present invention is to provide an echo cancellation technique for reducing the residual echo more than the conventional method by targeting the diffuse residual echo.

  In order to solve the above-described problem, according to the first aspect of the present invention, the echo canceller is configured such that P is an integer equal to or greater than 2, and P speakers and P microphones are arranged in a common sound field. Then, when the received signal is reproduced from the speaker, the echo that goes around the microphone via the echo path is deleted. The echo canceller estimates the diffuse residual echo contained in the sound signal collected in the wave number domain using the signal obtained by converting the sound signal collected by the microphone into the wave number domain and the received signal in the wave number domain. A wave number domain residual echo estimation canceling unit that cancels the diffuse residual echo estimated from the collected sound signal. The wave number domain diffuse residual echo estimation / erasing unit calculates a power spectrum matrix that is a P × P matrix using the received signal vector X, which is a P-dimensional vector having received signal for each wave number as an element, and its complex conjugate and transpose. Then, a compression input for calculating a cross spectrum matrix which is a P × P matrix using a complex conjugate and transpose of a sound pickup signal vector which is a P-dimensional vector having a sound pickup signal for each wave number as an element and a received signal vector X An input / output transfer characteristic matrix which is a P × P matrix having an estimated value of input / output transfer characteristics of the received signal and the collected sound signal as elements using an output correlation coefficient calculation unit, a power spectrum matrix and a cross spectrum matrix A compression input / output transfer characteristic estimator for obtaining a spread residual which is a P-dimensional vector obtained by multiplying the received signal vector X by an input / output transfer characteristic matrix and having an estimated value of a diffuse residual echo for each wave number as an element. Comprising a diffusion residual echo estimator for determining an echo vector, and a subtraction unit for obtaining a difference between the estimated value of the diffusion residual echo between the picked-up signal and the wavenumber region of wavenumbers region.

  In order to solve the above problems, according to a second aspect of the present invention, an echo canceller is configured such that P is an integer equal to or greater than 2, and P speakers and P microphones are arranged in a common sound field. Then, when the received signal is reproduced from the speaker, the echo that goes around the microphone via the echo path is deleted. The echo canceller estimates the diffuse residual echo contained in the sound signal collected in the wave number domain using the signal obtained by converting the sound signal collected by the microphone into the wave number domain and the received signal in the wave number domain. A wave number domain residual echo estimation canceling unit that cancels the diffuse residual echo estimated from the collected sound signal. The wave number domain diffusion residual echo estimation elimination unit sets P ′ <P, and uses a compression matrix W that is a P ′ × P matrix to obtain a received signal vector X that is a P-dimensional vector having received signals for each wave number as elements. The difference between the received signal vector X and the input dimension compression unit that compresses the compressed vector Z into the P′-dimensional compressed vector Z, the P-dimensional vector obtained by expanding the compressed vector Z with the complex conjugate transpose matrix of the compression matrix W, and Next, a power spectrum matrix which is a P ′ × P ′ matrix is calculated using a dimensional compression matrix updating unit for updating the compression matrix W, a compressed vector Z and its complex conjugate and transposition, and a sound collected signal for each wave number is calculated. A compression input / output correlation coefficient calculating unit that calculates a cross spectrum matrix that is a P × P ′ matrix by using a complex conjugate and transpose of a sound pickup signal vector that is a P-dimensional vector as an element and a compression vector Z; Spect Compressed input / output transfer characteristic estimator for obtaining an input / output transfer characteristic matrix which is a P × P ′ matrix having an estimated value of the input / output transfer characteristics of the received signal and the collected sound signal as elements using the matrix and the cross spectrum matrix A diffusion residual echo estimator that multiplies the compression vector Z by an input / output transfer characteristic matrix to obtain a diffusion residual echo vector that is a P-dimensional vector whose element is an estimated value of diffusion residual echo for each wave number; A subtractor for obtaining a difference between the collected sound signal and an estimated value of the diffuse residual echo in the wave number domain.

  In order to solve the above-described problem, according to a third aspect of the present invention, an echo canceling method is such that P is an integer equal to or greater than 2, and P speakers and P microphones are arranged in a common sound field. Then, when the received signal is reproduced from the speaker, the echo that goes around the microphone via the echo path is deleted. The echo cancellation method estimates the diffuse residual echo contained in the collected sound signal in the wave number domain using the signal obtained by converting the collected sound signal collected by the microphone into the wave number domain and the received signal in the wave number domain. A wave number domain diffuse residual echo estimation canceling step for canceling the diffuse residual echo estimated from the collected sound signal. The wave number domain diffusion residual echo estimation elimination step calculates a power spectrum matrix that is a P × P matrix using the received signal vector X, which is a P-dimensional vector whose elements are received signals for each wave number, and its complex conjugate and transpose. Then, a compression input for calculating a cross spectrum matrix which is a P × P matrix using a complex conjugate and transpose of a sound pickup signal vector which is a P-dimensional vector having a sound pickup signal for each wave number as an element and a received signal vector X An input / output transfer characteristic matrix which is a P × P matrix whose elements are estimated values of input / output transfer characteristics of the received signal and the collected sound signal using the output correlation coefficient calculating step, the power spectrum matrix and the cross spectrum matrix A compression input / output transfer characteristic estimation step for obtaining the received signal vector X by multiplying the input / output transfer characteristic matrix by an input / output transfer characteristic matrix, Comprising a diffusion residual echo estimation step of obtaining a diffusion residual echo vector is Torr, and a subtraction step of obtaining a difference between the estimated value of the diffusion residual echo between the picked-up signal and the wavenumber region of wavenumbers region.

  In order to solve the above-described problem, according to a fourth aspect of the present invention, an echo canceling method uses P as an integer of 2 or more, and P speakers and P microphones are arranged in a common sound field. Then, when the received signal is reproduced from the speaker, the echo that goes around the microphone via the echo path is deleted. The echo cancellation method estimates the diffuse residual echo contained in the collected sound signal in the wave number domain using the signal obtained by converting the collected sound signal collected by the microphone into the wave number domain and the received signal in the wave number domain. A wave number domain diffuse residual echo estimation canceling step for canceling the diffuse residual echo estimated from the collected sound signal. The wave number domain diffusion residual echo estimation erasure step sets P ′ <P, and uses a compression matrix W that is a P ′ × P matrix to obtain a received signal vector X that is a P-dimensional vector having received signals for each wave number as elements. The difference between the received signal vector X and the input dimension compression step for compressing the compressed vector Z into the P′-dimensional compressed vector Z, the P-dimensional vector obtained by expanding the compressed vector Z with the complex conjugate transpose matrix of the compression matrix W, and the received signal vector X is minimized. Then, a power spectrum matrix that is a P ′ × P ′ matrix is calculated using a dimension compression matrix update step for updating the compression matrix W, and the compression vector Z and its complex conjugate and transposition, and a sound collected signal for each wave number is calculated. A compression input / output correlation coefficient calculation unit for calculating a cross spectrum matrix that is a P × P ′ matrix using a complex conjugate and transpose of a sound pickup signal vector that is a P-dimensional vector as an element and a compression vector Z And an input / output transfer characteristic matrix which is a P × P ′ matrix whose elements are estimated values of the input / output transfer characteristics of the received signal and the collected sound signal using the power spectrum matrix and the cross spectrum matrix. Input / output transfer characteristic estimation step, and diffusion residual echo estimation for multiplying the compression vector Z by an input / output transfer characteristic matrix to obtain a diffuse residual echo vector which is a P-dimensional vector having an estimated value of diffuse residual echo for each wave number as an element And a subtracting step for obtaining a difference between the collected sound signal in the wave number domain and the estimated value of the diffuse residual echo in the wave number domain.

  According to the present invention, it is possible to reduce the residual echo more than the conventional method.

The figure which shows the example of arrangement | positioning of the echo cancellation apparatus in a multichannel communication conference system. Functional block diagram of the echo canceller 100 The figure which shows the processing flow of the echo cancellation apparatus. The functional block diagram of a wave number domain echo replica production | generation part. The figure for demonstrating frame composition. The functional block diagram of a residual echo cancellation part. The figure which shows the processing flow of a residual echo cancellation part. The functional block diagram of the wave number domain residual echo estimation elimination part. The figure which shows the processing flow of a wave number area | region residual echo estimation elimination part. The functional block diagram of the wave number area | region spreading | diffusion residual echo estimation elimination part. The figure which shows the processing flow of the wave number area | region spreading | diffusion residual echo estimation elimination part. The functional block diagram of a residual echo cancellation part at the time of using the wave number area | region spreading | diffusion residual echo estimation cancellation part independently. The functional block diagram of the wave number area | region spreading | diffusion residual echo estimation cancellation | release part at the time of using a wave number area | region spreading | diffusion residual echo estimation cancellation | release part independently. The figure for demonstrating the processing result of a conventional method. The figure for demonstrating the processing result in the modification of 1st embodiment.

  Hereinafter, embodiments of the present invention will be described. In the drawings used for the following description, constituent parts having the same function and steps for performing the same process are denoted by the same reference numerals, and redundant description is omitted. In the following description, the symbol “^” or the like used in the text should be described immediately above the character immediately before, but it is described immediately after the character due to restrictions on text notation. In the formula, these symbols are written in their original positions. Further, the processing performed for each element of a vector or matrix is applied to all elements of the vector or matrix unless otherwise specified.

<Points of first embodiment>
In the first embodiment, there are provided means for estimating the transfer characteristic from the received signal to the diffuse residual echo in the wave number domain with high accuracy and low computational complexity, and means for subtracting the diffuse residual echo from the error signal in the wave number domain. The estimation of the diffuse residual echo is improved by estimating the transfer characteristic from the received signal in the wave number domain to the error signal in the wave number domain as a matrix. Furthermore, the amount of calculation is greatly reduced by compressing the received signal in the wave number domain and then using it for estimation. By utilizing the correlation between the compressed received signal and the error signal, the estimated fluctuation due to a signal other than the residual echo is suppressed.

<Echo Canceling Device 100 according to First Embodiment>
FIG. 1 shows an arrangement example of the echo cancellation apparatus 100 in the multi-channel communication conference system, FIG. 2 shows a functional block diagram of the echo cancellation apparatus 100, and FIG. 3 shows a processing flow thereof.
The multi-channel communication conference system including the echo canceller 100 includes a P-channel playback system and a P-channel sound collection system. However, P ≧ 2. In this multi-channel communication conference system, P speakers 2 _p and P microphones 3 _p are arranged in a common sound field. P-channel of the received signal x (p, n) is reproduced as an acoustic signal by the loudspeaker 2 _p, wraps around each of the P number of microphones 3 _p through the acoustic echo path. This signal component that wraps around is the aforementioned echo. Here, p = 1, 2,..., P, and n is an index representing time.

Echo canceller 100 receives the received signal x (p, n) via the respective P-number of the receiving end _{1 p,} collected signal y (p being picked up by each of the P number of microphones _{3 p,} n ). Furthermore, to erase the echoes from each of the P number of collected signal y (p, n), and generates a transmission signal ^{e (3) (p, n} ), and outputs the transmission terminal 4 _p.

The echo cancellation apparatus 100 includes a frequency domain conversion unit 11, a wave number conversion unit 12, a wave number domain echo replica generation unit 21, an inverse wave number conversion unit 31, a time domain conversion unit 32, a frame synthesis unit 34, and P pieces. of including a subtraction unit 33 _p, and the error frequency domain transform section 41, and an error-wavenumber conversion unit 42. Note that the echo cancellation apparatus 100 implements a wavenumber domain adaptive algorithm using existing technology (see, for example, Non-Patent Document 2).
Further, echo canceling apparatus 100 includes a residual echo canceling unit 120 that estimates a residual echo from the received signal and the error signal in the wave number domain and subtracts the residual echo from the error signal. Details of each part will be described below.

<Frequency domain converter 11>
The frequency domain transform unit 11 receives the received signal x (p, n) in the time domain of the P channel, converts it into a received signal X _f (p, i) in the frequency domain for each channel p (s1), and P × 2F The received signal X _f (p, i) in the frequency domain is output to the wave number converter 12. However, i represents a frame number, 2F represents the number of samples included in one frame, f represents a frequency index, and f = 0, 1,..., 2F-1. If the sampling frequency of the signal is f _S , X _f (p, i) represents a component of the frequency f _S f / 2F [Hz] of the received signal of channel p in frame i. As a method of frequency domain transformation, Fast Fourier Transform (hereinafter abbreviated as “FFT”) or the like can be considered.

First, the frequency domain transform unit 11 receives 2F received signals x (p every time F / D received signals x (p, n) are received (in other words, every time n = iF / D). , N−2F + 1), x (p, n−2F + 2),..., X (p, n) are blocked for one frame to obtain a received signal x (p, i) in units of frames. However, F is a natural number and D is a natural number that divides F. For example,
x (p, i) = [x (p, (iF / D) -2F + 1), x (p, (iF / D) -2F + 2), ..., x (p, iF / D)] ^T (1)
It is. However, ^T represents transposition. Hereinafter, each signal is blocked by 1 frame = 2F samples and shift amount F / D samples. In order to simplify and speed up the FFT calculation, F is often raised to a power of 2. Hereinafter, a case of D ≧ 2 is shown.

Further, the frequency domain converting unit 11 converts the received signal x (p, i) in units of frames into a received signal X (p, i) in the frequency domain as shown in the following equation.
X (p, i) = FFT (x (p, i)) = [X ₀ (p, i)… X _f (p, i)… X _2F-1 (p, i)] (2)
Note that each signal in the frequency domain, including the received signal X (p, i), is represented by a short-time spectrum.

<Wave number converter 12>
The wave number converter 12 receives P × 2F frequency domain received signals X _f (p, i), and receives the received signal X ^{(W (W} ) in the wave number domain for each frequency f according to the following equations (3) and (4). ⁾ _F (k, i) is converted (s3), and P × 2F reception signals X ^(W) _f (k, i) in the wave number domain are output to the wave number domain echo replica generation unit 21 and the residual echo cancellation unit 120 To do. Where k is a wave number index, K is a natural number, K = −K + 1, −K + 2,..., −1, 0, 1,. When the number P is an odd number and P = 2K + 1, k = −K, −K + 1,..., −1, 0, 1,.

(1) When the number of channels P is an even number and P = 2K,
X ^(W) _f (i) = FFT ([X _f (1, i) X _f (2, i)… X _f (P, i)])
= [X ^(W) _f (0, i)… X ^(W) _f (k, i)… X ^(W) _f (K, i) X ^(W) _f (-K + 1, i)… X ^{( W)} _f (-1, i)]
(3)
It is.
(2) When the number of channels P is odd and P = 2K + 1,
X ^(W) _f (i) = FFT ([X _f (1, i) X _f (2, i)… X _f (P, i)])
= [X ^(W) _f (0, i)… X ^(W) _f (k, i)… X ^(W) _f (K, i) X ^(W) _f (-K, i)… X ^(W) _f (-1, i)] (4)
It is. Since the conversion to the wave number domain is performed at high speed with an FFT having a power of 2, the following description will be given for the case where the number of channels P is an even number (P = 2K). Each signal in the wave number domain including the received signal X ^(W) _f (k, i) is represented by a short-time spectrum.

<Wave number domain echo replica generator 21>
The wave number domain echo replica generation unit 21 receives the received signal X ^(W) _f (k, i) in the P × 2F wave number domain and the error signal E ^(W) _f (k, i) in the P × 2F wave number domain. (Details will be described later), and using these values, P × (F + 1) number of wave number domain echo replicas Y ^ ^(W) _f (k, i) are generated for f ≦ F and vice versa. Output to the wave number converter 31. The echo replica imitates an echo included in the collected sound signal and is an estimated value of the echo.

  FIG. 4 shows a functional block diagram of the wave number domain echo replica generation unit 21. The wave number domain echo replica generation unit 21 includes a correction amount calculation unit 211, a filter coefficient unit 213, and a multiplication unit 215.

(Multiplier 215)
The multiplication unit 215 of the wave number domain echo replica generation unit 21 receives the received signal X ^(W) _f (k, i) of P × 2F wave number regions. Also, P × (F + 1) × (2δ + 1) filter coefficients H ^(W) _f (k, k + dk, i) (where f ≦ F) are received from a filter coefficient unit 213 described later. However, dk = −δ, −δ + 1,..., −1, 0, 1,. As δ, 1 or 2 is recommended in Non-Patent Document 2. The multiplication unit 215 multiplies the received signal X ^(W) _f (k, i) by the filter coefficient H ^(W) _f (k, k + dk, i) in the wave number domain when f ≦ F. The echo replica Y ^ ^(W) _f (k, i) is generated (s5) and output to the inverse wave number converter 31.

By generating the echo replica Y ^ ^(W) _f (k, i) in the wave number domain in this way, adjacent spatial frequency components can be included. If it is not necessary to include adjacent spatial frequency components, an echo replica Y ^ ^(W) _f (k, i) in the wave number domain may be generated by the following equation with δ = 0.
Y ^ ^(W) _f (k, i) = H ^(W) _f (k, k, i) X ^(W) _f (k, i) (6)
The processing of the correction amount calculation unit 211 and the filter coefficient unit 213 will be described later.

<Reverse Wave Number Converter 31>
The inverse wave number conversion unit 31 receives echo replicas Y ^ ^(W) _f (k, i) of P × (F + 1) wave number regions (where f ≦ F), and the frequency region for each frequency f as in the following equation: Is converted to an echo replica Y ^ _f (p, i) (S9).
[Y ^ _f (1, i) Y ^ _f (2, i)… Y ^ _f (P, i)]
= IFFT ([Y ^ ^(W) _f (0, i)… Y ^ ^(W) _f (k, i)… Y ^ ^(W) _f (K, i) Y ^ ^(W) _f (-K + 1 , i)… Y ^ ^(W) _f (-1, i)])
(7)
For the frequency f> F, the echo replica Y ^ _f (p, i) in the frequency domain is obtained by the following equation from the symmetry regarding the FFT result of the real signal.
Y ^ _f (p, i) = conj (Y ^ _2F-f (p, i)) (8)
Here, conj (·) means taking a complex conjugate of •. The total P × 2F frequency domain echo replicas ＾ _f (p, i) obtained in this way are output to the time domain transform unit 32. As the inverse wave number conversion method, a method corresponding to the wave number domain conversion method in the wave number conversion unit 12 may be used.

<Time domain conversion unit 32>
Time domain transforming section 32 receives the echo replica _Y ^ f of P × 2F frequency-domain (p, i), the following equation, an echo replica in the frequency domain for each channel p _Y ^ f (p, i ) Is subjected to inverse FFT and converted to an echo replica signal vector y ^ (p, i) (the number of elements is F) in the time domain (s9).
y ^ (p, i) = [I _F 0 _F ] IFFT ([Y ^ ₀ (p, i)… Y ^ _f (p, i)… Y ^ _2F-1 (p, i)]) (9)
Here, 0 _F is an F × F zero matrix, and _IF is an F × F unit matrix. P time echo replica signal vectors y ^ (p, i) are output to the frame synthesis unit 34. As the time domain conversion method, a method corresponding to the frequency domain conversion method in the frequency domain conversion unit 11 may be used.

<Frame synthesis unit 34>
The frame synthesizer 34 receives P time-domain echo replica signal vectors y ^ (p, i). When the received signal x (p, n) is framed with D ≧ 2 in the frequency domain transform unit 11, the frame synthesis unit 34 matches the echo replica signal vector y ^ (p, i) obtained in the frame i. A windowing process is performed on the echo replica signal vector y ^ (p, i-1) obtained in the previous frame i-1, and then synthesized (s13), and P time domains after synthesis are performed. The echo replica signal vectors y ^ '(p, i) are output to the P subtracting units 33 _p .

In the case of D = 2, the Hanning window of length F / D is set to _WH , and the synthesized echo replica signal vector y ^ '(p, i) of length F / D is calculated by the following equation. The state of this synthesis is shown in FIG.
y ^ '(p, i-1) = [0 _{F / D} I _{F / D} ] diag (W _H ) y ^ (p, i-1) + [I _{F / D} 0 _{F / D} ] diag (W _H ) y ^ (p, i) (10)
_{However, 0 F / D} is zero _{matrix, I F / D} is a unit matrix of (F / D) × (F / D), diag (·) is a-pair (F / D) × (F / D) The matrix is a corner component and the others are zero.

<Subtraction unit 33 _p >
The subtractor 33 _p receives the time-domain echo replica signal vector y ^ ′ (p, i−1) and the collected sound signal y (p, n) collected by the microphone 3 _p . The echo replica signal is F / D delayed for frame synthesis. Taking this into consideration, the collected sound signal y (p, n) is 1 frame = F samples, and the shift amount is F / D samples.
y (p, i-1) = (y (p, ((i-1) F / D) -F + 1), y (p, ((i-1) F / D) -F + 2), …, Y (p, (i-1) F / D)] ^T
And the collected sound signal vector y (p, i-1). The subtractor 33 _p subtracts the echo replica signal vector y ^ '(p, i-1) in the time domain from the collected signal vector y (p, i-1) in the time domain as in the following equation (s11), A time domain error signal vector e (p, i) (the number of elements is F) is obtained and output to the residual echo canceling unit 120 and the error frequency domain converting unit 41.
e (p, i) = y (p, i-1) -y ^ '(p, i-1) (11)
With such a configuration, the echo canceller 100 attempts to cancel echo.

<Error frequency domain conversion unit 41>
The error frequency domain transform unit 41 receives P time domain error signal vectors e (p, i), and sets the time domain error signal vector e (p, i) to 0 for each channel p as shown in the following equation. The padding is converted into the frequency domain (s15), and P × 2F frequency domain error signals E _f (p, i) are output to the error wave number converter 42.

<Error wave number converter 42>
The error wave number conversion unit 42 receives P × 2F frequency domain error signals E _f (p, i), and uses the following equation to calculate the wave number domain error signal E ^(W) _f (k, i) for each frequency f. (S17), and P × 2F wave number domain error signals E ^(W) _f (k, i) are output to the wave number domain echo replica generation unit 21.
E ^(W) _f (p, i) = FFT ([E _f (1, i)… E _f (P, i)]
= [E ^(W) _f (0, i)… E ^(W) _f (k, i)… E ^(W) _f (K, i) E ^(W) _f (-K + 1, i)… E ^{( W)} _f (-1, i)]
(13)

(Correction amount calculation unit 211)
The correction amount calculation unit 211 in the wave number domain echo replica generation unit 21 receives the received signal X ^(W) _f (k, i) in the P × 2F wave number domain and the error signal E ^{(W in the} P × 2F wave number domain). ⁾ _F (k, i) is received (see FIG. 2 and FIG. 4), and in f (f ≦ F), the filter coefficient of the adaptive filter in the wavenumber domain in the range of −K + 1 ≦ k ≦ K Correction amount dH ^(W) _f (k, k + dk, i) (where −δ ≦ dk ≦ δ) is calculated (s19), and P × (F + 1) × (2δ + 1) correction amounts dH ^(W) _f ( k, k + dk, i) is output to the filter coefficient unit 213.

なお、ρは分母が０になることを防止するための微小な正定数であり、右辺分母中のＢ^（Ｗ） _ｆ（ｋ，ｉ）は修正量ｄＨ^（Ｗ） _ｆ（ｋ，ｋ＋ｄｋ，ｉ）を補正しており、

Note that ρ is a minute positive constant for preventing the denominator from becoming 0, and B ^(W) _f (k, i) in the right-side denominator is the correction amount dH ^(W) _f (k, k + dk, i). )

により計算される。Ｂ^（Ｗ） _ｆ（ｋ，ｉ）は受話信号Ｘ^（Ｗ） _ｆ（ｋ−δ，ｉ）〜Ｘ^（Ｗ） _ｆ（ｋ＋δ，ｉ）のパワーの総和であり、βはパワー計算で短時間平均をとるための平滑化定数であり、０〜１の値をとる。

Is calculated by B ^(W) _f (k, i) is the sum of the powers of the received signals X ^(W) _f (k−δ, i) to X ^(W) _f (k + δ, i), and β is a short time in the power calculation. This is a smoothing constant for taking an average and takes a value of 0 to 1.

(Filter coefficient part 213)
The filter coefficient unit 213 in the wave number domain echo replica generation unit 21 receives P × (F + 1) × (2δ + 1) correction amounts dH ^(W) _f (k, k + dk, i) (where f ≦ F), The filter coefficient H ^(W) _f (k, k + dk, i) is updated by the equation (s21), and P × (F + 1) × (2δ + 1) wave number domain filter coefficients H ^(W) _f (k, k + dk, i + 1) is output to the multiplier 215.
H ^(W) _f (k, k + dk, i + 1) = H ^(W) _f (k, k + dk, i) + μdH ^(W) _f (k, k + dk, i) (16)
However, μ is a step size taking a value of 0-1. The processing in the multiplication unit 215 is as described above.

<Residual echo canceller 120>
The residual echo canceling unit 120 receives the received signal X ^(W) _f (k, i) in the P × 2F wave number domain and the error signal vector e (p, i) in the P time domain, and receives the wave number domain. Is estimated, the residual echo estimated from the error signal in the wave number domain is deleted (s23), and P time transmission signals e ⁽³⁾ (p, n) are output. .

  FIG. 6 is a functional block diagram of the residual echo canceling unit 120, and FIG. 7 shows its processing flow. The residual echo cancellation unit 120 includes a frequency domain conversion unit 121, a wave number conversion unit 122, a wave number domain residual echo estimation cancellation unit 1231, a wave number domain diffuse residual echo estimation cancellation unit 1232, an inverse wave number conversion unit 124, a time domain A conversion unit 125 and a frame synthesis unit 126 are included. Residual echoes include those based on direct waves that do not depend on reflections, and those based on reflected waves other than direct waves (diffuse residual echoes). In the residual echo canceling unit 120, the residual echo due to the direct wave is estimated by the wave number domain residual echo estimation canceling unit 1231 and the diffuse residual echo is estimated by the wave number domain residual residual echo estimation canceling unit 1232 and erased. Details of the processing will be described below.

(Frequency domain transform unit 121)
The frequency domain transform unit 121 receives P time domain error signal vectors e (p, i), and one error signal vector e (p, i) in the frame i for each channel p as shown in the following equation. Using the error signal vector e (p, i−1) in the previous frame i−1, the error signal is converted into a frequency domain error signal E ⁽¹⁾ _f (p, i) (s231), and P × 2F The frequency domain error signal E ⁽¹⁾ _f (p, i) is output to the wave number converter 122. For example, conversion into the frequency domain is performed by a method similar to that of the frequency domain conversion unit 11.
E ⁽¹⁾ (p, i) = FFT ([e ^T (p, i-1), e ^T (p, i)]) = [E ⁽¹⁾ ₀ (p, i)… E ⁽¹⁾ _f (p, i)… E ⁽¹⁾ _2F-1 (p, i)]
(17)

(Wave number converter 122)
The wave number converter 12 receives P × 2F frequency domain error signals E ⁽¹⁾ _f (p, i), and calculates the wave number domain error signal E ^(W1) _f (k, k ^{) for} each frequency f according to the following equation. i) (s232), and P × 2F wave number domain error signals E ^(W1) _f (k, i) are output to the wave number domain residual echo estimation elimination section 1231.
E ^(W1) _f (i) = FFT ([E ⁽¹⁾ _f (1, i) E ⁽¹⁾ _f (2, i)… E ⁽¹⁾ _f (P, i)])
= [E ^(W1) _f (0, i)… E ^(W1) _f (k, i)… E ^(W1) _f (K, i) E ^(W1) _f (-K + 1, i)… E ^{( W1)} _f (-1, i)]
(18)

(Wave number domain residual echo estimation elimination unit 1231)
The wave number domain residual echo estimation elimination unit 1231 receives P × 2F wave number domain received signals X ^(W) _f (k, i−1) and P × 2F wave number domain error signals E ^(W1) _f ( k, i) and using these values, the residual echo due to the direct wave included in the error signal E ^(W1) _f (k, i) is estimated when f ≦ F, and the sound collected signal in the wave number domain is estimated. The residual echo due to the direct wave estimated from (1) is eliminated (s2331), and the error signal E ^(W2) _f (p, i) in the P × (F + 1) wave number domain is obtained by eliminating the residual echo due to the direct wave. Note that the immediately preceding X ^(W) _f (k, i−1) is used as the frequency domain received signal, not X ^(W) _f (k, i), when the echo replica signal is frame-synthesized. This is because the delay caused by the above is taken into consideration.

Details of the processing will be described below.
FIG. 8 is a functional block diagram of the wave number domain residual echo estimation erasing unit 1231, and FIG. 9 shows a processing flow thereof.
Wave number domain residual echo estimation elimination section 1231 includes an input / output correlation coefficient calculation section 12311, an input / output transfer characteristic estimation section 12312, a residual echo estimation section 12313, a residual echo correction section 12314, and a subtraction section 12315.

((Input / output correlation coefficient calculation unit 12311))
The input / output correlation coefficient calculation unit 12311 receives the received signal X ^(W) _f (k, i−1) in the P × 2F wave number domain and the error signal E ^(W1) _f (k in the P × 2F wave number domain ^). , I), and in order to estimate the transfer characteristics of the system that outputs the residual echo signal in the wave number domain at f ≦ F, the received signal X ^(W) _f (in the wave number domain at time n = iF / D k, i−1) and the error signal E ^(W1) _f (k, i) in the wave number domain.
P _f (k, i) = E [X ^{(W) *} _f (k, i-1) X ^(W) _f (k, i-1)]
Q _f (k, i) = E [X ^{(W) *} _f (k, i-1) E ^(W1) _f (k, i)] (19)
Thus, the power spectrum P _f (k, i) of the received signal and the cross spectrum Q _f (k, i) between the received signal and the error signal are calculated (s2331a), and the input / output transfer characteristic estimating unit 12312 Output. However, i is a frame number, and there is a relationship of n = iF / D with time n, * represents a complex conjugate, and E [•] represents an average of. As an example of the averaging process,
E [X ^{(W) *} _f (k, i-1) X ^(W) _f (k, i-1)] = βE [X ^{(W) *} _f (k, i-2) X ^(W) _f ( k, i-2)] + (1-β) X ^{(W) *} _f (k, i-1) X ^(W) _f (k, i-1)
As described above, a method using a processing result of one frame before and a smoothing constant β that takes a value of 0 to 1 or a method of obtaining a statistical average value of past several to several tens of frames can be considered.

((Input / output transfer characteristic estimation unit 12312))
The input / output transfer characteristic estimation unit 12312 receives P × (F + 1) power spectra P _f (k, i) and P × (F + 1) cross spectra Q _f (k, i), and receives f (f ≦ f F), from the power spectrum P _f (k, i) and the cross spectrum Q _f (k, i)

により、受話信号と誤差信号との入出力伝達特性を推定し（ｓ２３３１ｂ）、推定値Ｇ’_ｆ（ｋ，ｉ）を残留エコー推定部１２３１３に出力する。

Thus, the input / output transfer characteristics between the received signal and the error signal are estimated (s2331b), and the estimated value G ′ _f (k, i) is output to the residual echo estimator 12313.

Further, the estimated value G ′ _f (k, i) may be smoothed by the following equation, and the smoothed estimated value G _f (k, i) may be output to the residual echo estimating unit 12313.

本実施形態では、平滑化した推定値Ｇ_ｆ（ｋ，ｉ）を出力するものとする。ここで、β_２は、入出力伝達特性の推定値を平滑化するための定数であり、０〜１の間の値をとる。

In the present embodiment, it is assumed that a smoothed estimated value G _f (k, i) is output. Here, beta ₂ are constants for smoothing the estimate of the input-output transfer characteristic, it takes a value between 0 and 1.

((Residual Echo Estimator 12313))
The residual echo estimator 12313 includes P × (F + 1) wave number domain received signals X ^(W) _f (k, i−1), P × (F + 1) estimated values G _f (k, i), and Then, at f (f ≦ F), the received signal X ^(W) _f (k, i−1) is multiplied by the estimated value G _f (k, i) as shown in the following equation to estimate the residual echo. (S2331c) and the estimated value ΔE ^(W1) _f (k, i) are output to the residual echo correcting unit 12314.
ΔE ^(W1) _f (k, i) = G _f (k, i) X ^(W) _f (k, i-1) (21)

((Residual echo correction unit 12314))
The residual echo correcting unit 12314 generates P × (F + 1) estimated values ΔE ^(W1) _f (k, i) and P × 2F error signals E ^(W1) _f (k, i). Then, at f (f ≦ F), it is corrected by the following equation (s2331d), and the corrected residual echo estimated value ΔE ^{II (W1)} _f (k, i) is output to the subtracting unit 12315.

ただし、式中のＳ^（Ｗ） _ｆ（ｋ，ｉ）は、送話信号の推定値であり、次式により算出される。
S^(W) _f(k,i)=E^(W1) _f(k,i)-ΔE^(W1) _f(k,i) (23)
また、Ｔは各スペクトルの推定の自由度の数であり、入出力相関係数算出部１２３１１においてパワースペクトルＰ_ｆ（ｋ，ｉ）及びクロススペクトルＱ_ｆ（ｋ，ｉ）を算出するときのフレーム数が、これにあたる。Ｍは入力変数の数であり、式（２０）の場合にはＭ＝１になる。またＦ_{２Ｍ，Ｔ−２Ｍ，ａｌｐｈａ}は、自由度ｎ_１＝２Ｍ、ｎ_２＝Ｔ−２ＭのＦ分布の１００×ａｌｐｈａ百分比点である。

However, S ^(W) _f (k, i) in a type _| formula is an estimated value of a transmission signal, and is calculated by following Formula.
S ^(W) _f (k, i) = E ^(W1) _f (k, i) -ΔE ^(W1) _f (k, i) (23)
T is the number of degrees of freedom of estimation of each spectrum, and the frame when the input / output correlation coefficient calculation unit 12311 calculates the power spectrum P _f (k, i) and the cross spectrum Q _f (k, i). This is the number. M is the number of input variables. In the case of equation (20), M = 1. F _{2M, T-2M, and alpha} are 100 × alpha percentage points of F distribution with n ₁ = 2M and n ₂ = T-2M degrees of freedom.

  The F distribution is a continuous probability distribution used in statistics. In analysis of variance, which is a method of statistical hypothesis testing, it is used to determine the effect / significance of each factor by breaking the variation in the observed data into error variation and the variation of each factor.

According to Reference Document 1, the confidence interval of the input / output transfer characteristic estimation value G _f (k, i) estimated by the input / output transfer characteristic estimation unit 12312 when M = 1 is a ratio from the true value.

の幅を持つ。
（参考文献１）Ｊ．Ｓ．ベンダット、Ａ．Ｇ．ピアソル、「ランダムデータの統計的処理」、培風館、１９７６年、ｐ．１９４〜１９７

With a width of
(Reference 1) J. Org. S. Vendat, A.M. G. Pearsol, “Statistical Processing of Random Data”, Baifukan, 1976, p. 194-197

  In the estimation of the input / output transfer characteristic estimation unit 12311 based on the short-time spectrum, it is easier to estimate the correlation between the transmission and the residual echo than originally, and there is a tendency to estimate the transfer characteristic higher. Based on this, the above correction uses the value of the lower end of the confidence interval of the residual echo as the residual echo correction value.

((Subtraction unit 12315))
The subtracting unit 12315 calculates the error signal E ^(W1) _f (k, i) in the P × 2F wave number domain and the estimated value ΔE ^{II (W1)} of the residual echo after correction in the P × (F + 1) wave number domain. _f (k, i) is received, and at f (f ≦ F), an estimated value ΔE ^{II (W1)} _{f of the} residual echo from the error signal E ^(W1) _f (k, i) in the wave number domain as in the following equation: (K, i) is subtracted (s2331e) to obtain a difference E ^(W2) _f (k, i), which is output to the wave number domain diffuse residual echo estimation elimination unit 1232.
E ^(W2) _f (k, i) = E ^(W1) _f (k, i) -ΔE ^{II (W1)} _f (k, i) (25)
The difference E ^(W2) _f (k, i) is a signal obtained by eliminating the residual echo due to the direct wave from the error signal E ^(W1) _f (k, i), and the error signal E ^(W2) _f (k, i) It is also called i).

(Wave domain diffuse residual echo estimation elimination unit 1232)
Wave number domain diffuse residual echo estimation elimination section 1232 receives P × 2F received signal X ^(W) _f (k, i−2) in the wave number domain and P × (F + 1) wave number error signal E ^{(W2). )} _F (k, i) is received, and using these values, the diffuse residual echo included in the error signal E ^(W2) _f (k, i) is estimated when f ≦ F, and the error signal in the wavenumber domain is estimated. The diffuse residual echo estimated from E ^(W2) _f (k, i) is eliminated, and the transmission signal E ^(W3) _f (p, i) of P × (F + 1) wavenumber regions is obtained (s2332), and the inverse Output to wave number converter 124.

The wave number domain residual echo estimation cancellation unit 1232 uses (1) the reception signal X ^(W) _f (k, i−2) one frame before the wave number domain residual echo estimation cancellation unit 1231, and (2) the reception signal. X ^(W) _f (k, i−2) is also referred to as a vector (hereinafter also referred to as a wavenumber domain received signal vector)
X ^(W) _f (i-2) = [X ^(W) _f (0, i-2)… X ^(W) _f (k, i-2)… X ^(W) _f (K, i-2) X ^(W) _f (-K + 1, i-2)… X ^(W) _f (-1, i-2)]
In this case, a diffuse residual echo reflected and diffused by a wall surface or the like is used as an estimation target. Details of the processing will be described below.

  FIG. 10 is a functional block diagram of the wave number domain diffuse residual echo estimation erasing unit 1232, and FIG.

  Wave number domain diffusion residual echo estimation elimination section 1232 includes input dimension compression section 12320, dimension compression matrix update section 12326, compression input / output correlation coefficient calculation section 12321, compression input / output transfer characteristic estimation section 12322, and diffusion residual echo. An estimation unit 12323, a diffuse residual echo correction unit 12324, and a subtraction unit 12325 are included.

((Input dimension compression unit 12320))
The input dimension compression unit 12320 receives (F + 1) P ′ × P compression matrices W _f (i−1) updated by a dimension compression matrix update unit 12326, which will be described later, and P × 2F wave number domain receptions. The signal X ^(W) _f (k, i−2) is received. The received signal X ^(W) _f (k, i−2) in the P × 2F wave number domain is treated as 2F wave number domain received signal vector X ^(W) _f (i−2). The input dimension compression unit 12320 uses the compression matrix W _f (i−1), and in f ≦ F, the wave number domain received signal vector X ^(W) _f (i−2) is converted into a P′-dimensional wave number domain compression vector. Compressed to Z ^(W) _f (i−2) (s2332a), and outputs the result to the compressed input / output correlation coefficient calculation unit 12321 and the dimension compression matrix update unit 12326.

Z ^(W) _f (i-2) = W _f (i-1) X ^(W) _f (i-2)
It should be noted that P ′ <P, and the size of P ′ may be appropriately set depending on the environment (for example, the size of the room and the degree of reverberation). It can be set to about 10.

((Dimension compression matrix update unit 12326))
The dimension compression matrix update unit 12326 includes (F + 1) wave number domain compression vectors Z ^(W) _f (i-2) and P × 2F wave number domain received signals X ^(W) _f (k, i−2). And receive. The received signal X ^(W) _f (k, i−2) in the P × 2F wave number domain is treated as 2F wave number domain received signal vector X ^(W) _f (i−2). The dimension compression matrix update unit 12326 converts the wave number domain compression vector Z ^(W) _f (i-2) to the complex conjugate transpose matrix W ^H _f (i-1) of the compression matrix W _f (i-1) when f ≦ F. To obtain a difference dX ^(W) _f (i-2) from the wave number domain received signal vector X ^(W) _f (i-2).・^H represents the complex conjugate and transpose of
dX ^(W) _f (i-2) = X ^(W) _f (i-2)-W ^H _f (i-1) Z ^(W) _f (i-2)
= X ^(W) _f (i-2)-W ^H _f (i-1) W _f (i-1) X ^(W) _f (i-2)
Then, the compression matrix W _f (i−1) is updated so that the magnitude of the difference dX ^(W) _f (i−2) is minimized (s2332g), and the updated compression matrix W _f (i) is input to the input dimension. The data is output to the compression unit 12320.

  For this update, for example, a subspace tracking method can be used. As an example, the details when using OPSA1 in Reference 2 are as follows.

An inverse matrix R ^-1 _ZZ (i-2) of the autocorrelation matrix R _ZZ (i-2) of the wave number domain compression vector Z ^(W) _f (i-2) is set to an initial value R ^-1 _ZZ (0) = δ. _It estimates repeatedly from ₀ ⁻¹ I. However, δ ₀ is a non-zero positive constant, and prevents 0% when the iterative estimation process is executed for the first time. I is a P ′ × P ′ identity matrix. K (i) is a P′-dimensional intermediate generation vector, and V (i) is a P-dimensional intermediate generation vector. λ is a forgetting constant that takes a value between 0 and 1, and is a parameter that determines the estimated speed. The compression matrix W _f (i) can be updated as follows.
k (i) = R ^-1 _ZZ (i-3) Z ^(W) (i-2) / {λ + Z ^{(W) H} (i-3) R ^-1 _ZZ (i-3) Z ^(W) (i-2)}
R ^-1 _ZZ (i-2) = (1 / λ) {R ^-1 _ZZ (i-3) -k (i) Z ^{(W) H} (i-2) R ^-1 _ZZ (i-3)}
V (i) = dX ^(W) _f (i-2)-0.5 || dX ^(W) _f (i-2) || ² W ^H _f (i-1) k (i)
W _f (i) = W _f (i-1) + k (i) V ^H (i) / {1 + 0.25 || dX ^(W) _f (i-2) || ² || k (i) | | ² }
(Reference 2) SC Douglas and X. Sun, "Designing orthonormal subspace tracking algorithms", the Thirty-Fourth Asilomar Conference on Signals, Systems and Computers 2000, 2000, vol. 2, pp. 1441--1445.

((Compressed input / output correlation coefficient calculation unit 12321))
The compression input / output correlation coefficient calculation unit 12321 includes (F + 1) wave number domain compression vectors Z ^(W) _f (i−2) and P × (F + 1) wave number domain error signals E ^(W2) _f (k , I). Note that the error signal E ^(W2) _f (k, i) in the P × (F + 1) wave number domain is changed to the (F + 1) wave number domain error signal vector E ^(W2) _f (k, i) (where E ^{( W2)} _f (i) = [E ^(W2) _f (0, i)… E ^(W2) _f (k, i)… E ^(W2) _f (K, i) E ^(W2) _f (-K + 1 , i)... E ^(W2) _f (-1, i)]) (where f ≦ F). The compression input / output correlation coefficient calculation unit 12321 has (F + 1) wavenumber domain compression vectors Z ^(W) _f (i-2) and (F + 1) wavenumber domain error signal vectors E ^(W2) when f ≦ F. and _{f (i)} receiving signals compressed from the power spectrum matrix _{^{P (2) f (i)}} , cross-spectral matrix Q between the compressed received signal and the error signal and ^{₍₂₎ f _(i)} It is calculated by the following equation (s2332b) and output to the compression input / output transfer characteristic estimation unit 12322.
P ⁽²⁾ _f (i) = E [Z ^(W) _f (i-2) Z ^{(W) H} _f (i-2)]
Q ⁽²⁾ _f (i) = E [E ^(W2) _f (i) Z ^{(W) H} _f (i-2)]

((Compression input / output transfer characteristic estimation unit 12322))
The compression input / output transfer characteristic estimation unit 12322 calculates a power spectrum matrix P ⁽²⁾ _f (i) which is a P ′ × P ′ matrix and a cross spectrum matrix Q ⁽²⁾ _f (i) which is a P × P ′ matrix. receive. Each matrix is (F + 1). The compression input / output transfer characteristic estimation unit 12322 calculates the input / output transfer from the power spectrum matrix P ⁽²⁾ _f (i) and the cross spectrum matrix Q ⁽²⁾ _f (i) according to the following equation at f (f ≦ F). A characteristic matrix G ′ _f (i) is obtained (s2332c) and output to the diffuse residual echo estimator 12323.

The input / output transfer characteristic matrix G ′ _f (i) is a P × P ′ matrix whose elements are estimated values of the input / output transfer characteristics of the compressed reception signal and error signal. In compression of the received signal, a wave number domain received signal vector (its elements are each wave number component) is decomposed into main components (main patterns) and approximated in a manner similar to principal component analysis. The correspondence between each main component and each wave number component of the residual echo is described by an input / output transfer characteristic matrix G ′ _f (i).

Further, the estimation matrix G ′ _f (i) may be smoothed by the following equation, and the smoothed input / output transfer characteristic matrix G _f (i) may be output to the diffuse residual echo estimation unit 12323.

In this embodiment, a smoothed input / output transfer characteristic matrix G _f (i) is output. Here, beta ₂ are constants for smoothing the estimate of the input-output transfer characteristic, it takes a value between 0 and 1.

((Diffusion residual echo estimation unit 12323))
The diffuse residual echo estimator 12323 receives (F + 1) wave number domain compression vectors Z ^(W) _f (i−2) and (F + 1) input / output transfer characteristic matrices G _f (i), and receives f ( In f ≦ F), the diffusion residual echo vector ΔE ^(W2) _{f is} obtained by multiplying the compression vector Z ^(W) _f (k, i−2) by the input / output transfer characteristic matrix G _f (i) as in the following equation. (I) is obtained (s2332d) and output to the diffusion residual echo correction unit 12324.
ΔE ^(W2) _f (i) = G _f (i) Z ^(W) _f (i-2)
The diffuse residual echo vector ΔE ^(W2) _f (i) is a P-dimensional vector whose element is an estimated value of diffuse residual echo for each wave number.

((Diffusion residual echo correcting unit 12324))
The diffusion residual echo correction unit 12324 includes (F + 1) diffusion residual echo vectors ΔE ^(W2) _f (i) and P × (F + 1) number of error signals E ^(W2) _f (k, i). In f (f ≦ F), each element ΔE ^(W2) _f (k, i) of the diffusion residual echo vector ΔE ^(W2) _f (i) is corrected by the following equation (s2332e), and the diffusion after the correction The estimated value ΔE ^{II (W2)} _f (k, i) of the residual echo is output to the subtracting unit 12325.

However, S ^(W2) _f (k, i) in a formula is an estimated value of a transmission signal, and is calculated by the following formula.
S ^(W2) _f (k, i) = E ^(W2) _f (k, i) -ΔE ^(W2) _f (k, i)
T is the number of degrees of freedom of estimation of each spectrum, and the compressed input / output correlation coefficient calculation unit 12321 calculates the power spectrum matrix P ⁽²⁾ _f (i) and the cross spectrum matrix Q ⁽²⁾ _f (i). This is the number of frames when calculating. M is the number of input variables. In the case of equation (30), M = 1. F _{2M, T-2M, and alpha} are 100 × alpha percentage points of F distribution with n ₁ = 2M and n ₂ = T-2M degrees of freedom.

((Subtraction unit 12325))
The subtracting unit 12325 calculates the error signal E ^(W2) _f (k, i) in the P × (F + 1) wave number domain and the estimated value ΔE ^II of the diffuse residual echo after the correction in the P × (F + 1) wave number domain. ^(W2) _f (k, i) is received, and at f (f ≦ F), an estimated value ΔE ^{II of the} diffuse residual echo from the error signal E ^(W2) _f (k, i) in the wave number domain as shown in the following equation: ^(W2) _f (k, i) is subtracted (s2332f), the difference is obtained as a transmission signal E ^(W3) _f (k, i) in the wave number domain, and is output to the inverse wave number converter 124.
E ^(W3) _f (k, i) = E ^(W2) _f (k, i) -ΔE ^{II (W2)} _f (k, i)

(Reverse wave number converter 124)
The inverse wave number converter 124 receives the transmission signal E ^(W3) _f (k, i) in the P × (F + 1) wave number domain (see FIG. 6), and at f (f ≦ F), For each frequency f, it is converted into a frequency domain transmission signal E ⁽³⁾ _f (p, i) (s234).
[E ⁽³⁾ _f (1, i) E ⁽³⁾ _f (2, i)… E ⁽³⁾ _f (P, i)]
= IFFT ([E ^(W3) _f (0, i)… E ^(W3) _f (k, i)… E ^(W3) _f (K, i) E ^(W3) _f (−K + 1, i)… E ^(W3) _f (-1, i)])
For the frequency f> F, the transmission signal E ⁽³⁾ _f (p, i) in the frequency domain is obtained by the following equation from the symmetry regarding the FFT result of the real signal.
E ⁽³⁾ _f (p, i) = conj (E ⁽³⁾ _2F-f (p, i))
The total P × 2F frequency domain transmission signals E ⁽³⁾ _f (p, i) thus obtained are output to the time domain conversion unit 125. As the inverse wave number conversion method, a method corresponding to the wave number domain conversion method in the wave number conversion unit 122 may be used.

(Time domain conversion unit 125)
Time domain transform section 125 receives the transmission signal _{^{E (3) f (p,}} i) of P × 2F frequency-domain, the following equation, transmission signal ^{E (3} in the frequency domain for each channel p ⁾ _F (p, i) is subjected to inverse FFT, converted into a time domain transmission signal vector e ⁽³⁾ (p, i) (number of elements is 2F) (s235), and output to the frame synthesis unit 126.
e ⁽³⁾ (p, i) = IFFT ([E ⁽³⁾ ₀ (p, i)… E ⁽³⁾ _f (p, i)… E ⁽³⁾ _2F-1 (p, i)])
As the time domain conversion method, a method corresponding to the frequency domain conversion method in the frequency domain conversion unit 121 may be used.

(Frame synthesis unit 126)
The frame synthesizing unit 126 receives P time domain transmission signal vectors e ⁽³⁾ (p, i). When the received signal x (p, n) is framed with D ≧ 2 in the frequency domain transform unit 121, the frame synthesizing unit 126 transmits the transmitted signal e ⁽³⁾ (p, i) obtained in the frame i. And the transmission signal e ⁽³⁾ (p, i-1) obtained in the previous frame i-1 are subjected to windowing processing, synthesized (s236), and the synthesized transmission Element e ⁽³⁾ (p, n−F / D + 1), e ⁽³⁾ (p, n−F / D + 2) of signal vector e ⁽³⁾ ′ (p, i) (number of elements is F / D) ,..., E ⁽³⁾ (p, n) are sequentially output as the output value of the echo canceller 100. However, there is a relationship of n = iF / D. The processing content is the same as the processing of the frame synthesis unit 34.

<Modification>
The residual echo canceling unit 120 can be used alone or as an echo canceling device. That is, in FIG. 2, the frequency domain transform unit 11, the wave number transform unit 12, the wave number domain echo replica generation unit 21, the inverse wave number transform unit 31, the time domain transform unit 32, the frame synthesis unit 34, the P subtraction units 33 _p , the error frequency A configuration in which the adaptive filter portion (also referred to as echo canceling portion) composed of the region converting portion 41 and the error wave number converting portion 42 is removed can also be used. In that case, the residual echo canceling unit 120 receives the collected sound signal y (p, n) instead of the error signal vector e (p, i), converts it into a vector, and performs the same processing.

  Further, the wave number domain residual echo estimation erasure unit 1231 can be used even if the residual echo correction unit 12314 is removed. Similarly, the wave number domain diffuse residual echo estimation erasure unit 1232 can be used even if the diffuse residual echo correction unit 12324 is removed. In that case, each subtraction unit receives a signal before correction and performs the same processing.

In the residual echo canceling unit 120, the wave number domain residual echo estimation canceling unit 1231 can be removed and the wave number domain diffuse residual echo estimating canceling unit 1232 can be used alone. In this case, as shown in FIGS. 12 and 13, the input of the wave number domain diffuse residual echo estimation erasure unit 1232 changes. FIG. 12 is a functional block diagram of the residual echo canceling unit 120 when the wave number domain diffuse residual echo estimation canceling unit 1232 is used alone, and FIG. 13 is a functional block diagram of the wave number domain residual residual echo estimating canceling unit 1232. The receiver side signal is received signal X ^(W) _f (k, i−1) from P × 2F wave number domain to received signal X ^(W) _f (k, i−1) of P × 2F wave number domain. Changes to. In addition, since there is no wave number domain residual echo estimation elimination section 1231, the error signal becomes error signal E ^(W2) _f (k, i) = E ^(W1) _f (k, i) of P × 2F wave number domain. In this configuration, when the frame length is increased, the error signal E ^(W1) _f (k, i) includes both the direct component and the reflection component of the received signal X ^(W) _f (k, i-1). It is valid.

Further, the echo canceller and the wavenumber domain residual echo estimate canceler 1231 may be removed. In that case, instead of the error signal E ^(W2) _f (k, i) of the error signal P × 2F wave number domain, the collected sound signal y (p, n) is received, and the collected signal Y ⁽ p) of the wave number domain. ^W) Convert to _f (k, i) and perform similar processing.

  As a method for obtaining an echo replica in the wave number domain, an existing technique other than the above-described method may be used. In addition, an echo replica may be obtained in the frequency domain or the time domain using existing technology. However, it is known that subtracting the time-domain echo replica from the time-domain sound pickup signal has higher echo cancellation accuracy, so even if the echo replica is obtained in the frequency domain, it is converted to the time domain. In addition, it is desirable to subtract from the time domain sound pickup signal.

  In the first embodiment, the case where the number of channels P is an even number has been described, but an odd number (P = 2K + 1) may be used.

In the present embodiment, the input dimension compression unit 12320 compresses the wave number domain received signal vector X ^(W) _f (i-2) into a wave number domain compressed vector Z ^(W) _f (i-2). However, it is not always necessary to compress. In that case, in the processing after the input dimension compression unit 12320, the wave number domain received signal vector X ^(W) _f (i-2) may be used instead of the wave number domain compressed vector Z ^(W) _f (i-2). . For example, the compression input / output correlation coefficient calculation unit 12321 obtains the power spectrum matrix P ⁽²⁾ _f (i) and the cross spectrum matrix Q ⁽²⁾ _f (i) by the following equations and the following equations, respectively.
P ⁽²⁾ _f (i) = E [X ^(W) _f (i-2) X ^{(W) H} _f (i-2)]
Q ⁽²⁾ _f (i) = E [E ^(W2) _f (i) X ^{(W) H} _f (i-2)]
In this case, the input dimension compression unit 12320 and the dimension compression matrix update unit 12326 may be removed. Alternatively, the processing of the dimension compression matrix update unit 12326 may be removed, and the input dimension compression unit 12320 may use a P × P unit matrix instead of the compression matrix W _f (i−1). Even in such a configuration, the residual echo can be reduced more than the conventional method in consideration of reflection of the wall surface or the like.

<Effect>
In the conventional method, the transfer characteristic from the received signal X ^(W) _f (i) in the wave number domain to the error signal E ^(W1) _f (i) in the wave number domain is estimated as a diagonal matrix, and residual echo cancellation is performed. This corresponds to estimating the residual echo considering only the direct propagation of the wavefront.

In this configuration, the transfer characteristic from the received signal X ^(W) _f (i) in the wave number domain to the error signal E ^(W2) _f (i) in the wave number domain is estimated as a matrix, and the diffusion residual echo vector in the wave number domain is calculated. Estimate and subtract from wave number domain error signal vector E ^(W1) _f (i). This corresponds to estimating the residual echo in consideration of the arrival of the wavefront reflected on the ceiling or wall.

  As a result, even if the echo path estimation and cancellation by the adaptive filter in the wave number domain is not sufficient, the residual echo can be reduced more quickly than the conventional method taking into account the reflection of the wall surface etc. regardless of the conversation state. There is an effect.

Furthermore, by performing dimensional compression of the received signal, it is possible to reduce the amount of memory and the amount of calculation required for the residual echo estimation. Amount of memory required to store the correlation matrix of the received signals is proportional to the square of the dimension, when compressing input dimension to a times (0 <a <1), can be compressed amount of memory a ^2. Also requires a calculation amount proportional to the cube of the dimensions in the inverse matrix calculation in the residual echo transfer characteristic estimate, the input dimension if compressed to a times (0 <a <1), the amount of computation in a ³ It can be compressed.

<Simulation results>
In order to verify the effect of residual echo cancellation, a simulation was performed on the configuration of the modified example.
As the configuration of the echo canceller 100, only the residual echo canceler 120 is used. Further, the internal wave number domain residual echo estimation elimination unit 1231 is removed, and the diffusion domain residual echo correction unit 12324 is removed in the wave number domain residual residual echo estimation elimination unit 1232. The wave number domain residual echo canceling unit 1232 is set to compress the received signal to ¼. Β = 0.98 as a smoothing constant for correlation calculation, λ = 0.1 as a forgetting constant for calculating an inverse matrix of a compression vector correlation matrix, and β ₂ = 0.1 for estimation of an estimated input / output transfer characteristic. I used it.

  As a conventional method compared with this, the method proposed in Non-Patent Document 3 was used. In the configuration, only the residual echo cancellation unit 120 is used as the configuration of the echo cancellation apparatus 100, and only the wave number domain residual echo estimation cancellation unit 1231 is used therein. The residual echo correction unit 12314 was removed.

  In order to generate a signal for use in the simulation, a linear speaker array (32 elements, spacing 6 cm) and a linear microphone array (32 elements, spacing 6 cm) are placed 50 cm apart in parallel in a room with a reverberation time of 150 ms (P = 32), the total echo path impulse response between the speaker and the microphone was measured. The sampling frequency fs was set to 8 kHz, and 2F = 1024 was used as the frame length. The received signal was generated by simulating the situation in which two sound sources arranged at different positions reproduce white noise alternately and simulated sound collection by 32 microphones.

  14 and 15 show the simulation results. FIG. 14 shows the processing result of the conventional method, and FIG. 15 shows the processing result of the modification of this embodiment. In both cases, the echo cancellation amount (ERLE) by the residual echo cancellation processing is plotted for odd-numbered channels out of 32 channels.

  As shown in FIG. 14, the ERLE of the conventional method stays on average only over 10 dB, whereas the ERLE of the proposed method averages over 20 dB on average from FIG. This shows that the proposed method effectively cancels the echo.

<Other variations>
The present invention is not limited to the above-described embodiments and modifications. For example, 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. In addition, it can change suitably in the range which does not deviate from the meaning of this invention.

<Program and recording medium>
In addition, various processing functions in each device described in the above embodiments and modifications may be realized by a computer. In that case, 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. Further, 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 storage unit. When executing the process, this computer reads the program stored in its own storage unit and executes the process according to the read program. As another embodiment of this program, a computer may read a program directly from a portable recording medium and execute processing according to the program. Further, each time a program is transferred from the server computer to the computer, 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 includes information provided for processing by the electronic computer and equivalent 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 addition, although each device is configured by executing a predetermined program on a computer, at least a part of these processing contents may be realized by hardware.

Claims (9)

E is an echo that eliminates an echo that goes around the microphone via an echo path when P is an integer of 2 or more, P speakers and P microphones are arranged in a common sound field, and a received signal is reproduced from the speakers An erasing device,
Using the signal obtained by converting the collected sound signal collected by the microphone into the wave number domain and the received signal in the wave number domain, the diffusion residual echo included in the collected signal in the wave number domain is estimated, and the wave number domain Including a wave number domain diffuse residual echo estimation canceling unit that cancels the diffuse residual echo estimated from the collected sound signal,
The wave number domain diffuse residual echo estimation erasure unit is
A power spectrum matrix, which is a P × P matrix, is calculated using a received signal vector X, which is a P-dimensional vector having the received signal for each wave number as an element, and its complex conjugate and transpose, and the collected sound signal for each wave number A compressed input / output correlation coefficient calculation unit that calculates a cross spectrum matrix that is a P × P matrix using a sound pickup signal vector that is a P-dimensional vector having elements as elements and a complex conjugate and transpose of the received signal vector X; ,
Using the power spectrum matrix and the cross spectrum matrix, compression input to obtain an input / output transfer characteristic matrix which is a P × P matrix having an estimated value of the input / output transfer characteristics of the received signal and the collected sound signal as elements. An output transfer characteristic estimator;
A spread residual echo estimator that multiplies the received signal vector X by the input / output transfer characteristic matrix to obtain a diffuse residual echo vector that is a P-dimensional vector whose elements are estimated values of the diffuse residual echo for each wave number;
A subtractor that obtains a difference between the sound collection signal in the wave number domain and the estimated value of the diffuse residual echo in the wave number domain,
Echo canceler. E is an echo that eliminates an echo that goes around the microphone via an echo path when P is an integer of 2 or more, P speakers and P microphones are arranged in a common sound field, and a received signal is reproduced from the speakers An erasing device,
Using the signal obtained by converting the collected sound signal collected by the microphone into the wave number domain and the received signal in the wave number domain, the diffusion residual echo included in the collected signal in the wave number domain is estimated, and the wave number domain Including a wave number domain diffuse residual echo estimation canceling unit that cancels the diffuse residual echo estimated from the collected sound signal,
The wave number domain diffuse residual echo estimation erasure unit is
P ′ <P, and using a compression matrix W that is a P ′ × P matrix, a received signal vector X that is a P-dimensional vector having the received signal for each wave number as an element is converted into a P′-dimensional compressed vector Z. An input dimension compression unit for compression;
A dimension compression matrix updating unit that updates the compression matrix W so that a difference between a P-dimensional vector obtained by expanding the compression vector Z with a complex conjugate transpose matrix of the compression matrix W and the received signal vector X is minimized. When,
A power spectrum matrix which is a P ′ × P ′ matrix is calculated using the compression vector Z and its complex conjugate and transpose, and a sound collection signal vector which is a P-dimensional vector having the sound collection signal for each wave number as an element. And a compressed input / output correlation coefficient calculating unit that calculates a cross spectrum matrix that is a P × P ′ matrix using the complex conjugate and transpose of the compressed vector Z;
Compression using the power spectrum matrix and the cross spectrum matrix to obtain an input / output transfer characteristic matrix that is a P × P ′ matrix whose elements are estimated values of input / output transfer characteristics of the received signal and the collected sound signal An input / output transfer characteristic estimation unit;
A diffusion residual echo estimator that multiplies the compression vector Z by the input / output transfer characteristic matrix to obtain a diffusion residual echo vector that is a P-dimensional vector whose elements are estimated values of the diffusion residual echo for each wave number;
A subtractor that obtains a difference between the sound collection signal in the wave number domain and the estimated value of the diffuse residual echo in the wave number domain,
Echo canceler. The echo canceller of claim 2,
P ⁽²⁾ _f is a power spectrum matrix, Q ⁽²⁾ _f is a cross spectrum matrix, Z ^(W) _f is a compression vector Z, E ^(W2) _f is a collected signal vector, and ^H is a complex of Conjugate and transpose, E [·] represents the average of ·, the compressed input / output correlation coefficient calculation unit calculates the power spectrum matrix by the following equation,
P ⁽²⁾ _f = E [Z ^(W) _f Z ^{(W) H} _f ]
The cross spectrum matrix is calculated by the following equation:
Q ⁽²⁾ _f = E [E ^(W2) _f Z ^{(W) H} _f ]
β ₂ is a constant for smoothing the estimated value of the input / output transfer characteristic, and the compressed input / output transfer characteristic estimation unit obtains the input / output transfer characteristic matrix by the following equation or the following equation:

Echo canceler. The echo canceller according to any one of claims 1 to 3,
Using the received signal in the wave number domain and the collected sound signal in the wave number domain, a residual echo due to a direct wave included in the collected sound signal in the wave number domain is estimated, and the direct wave estimated from the collected sound signal in the wave number domain A wave number domain residual echo estimation canceling unit that cancels residual echo due to
The wave number domain residual echo estimation elimination part
I / O correlation coefficient for calculating a power spectrum of the received signal and a cross spectrum between the received signal and the collected sound signal using the received signal in the wave number domain and the collected sound signal in the wave number domain A calculation unit;
Using the power spectrum and the cross spectrum, an input / output transfer characteristic estimation unit that estimates an input / output transfer characteristic of the received signal and the collected sound signal;
A residual echo estimator for multiplying the received signal in the wave number domain by the estimated value of the input / output transfer characteristic to estimate the residual echo in the wave number domain;
A second subtracting unit for obtaining a difference between the sound pickup signal in the wave number region and the estimated value of the residual echo in the wave number region;
The collected sound signal used in the wave number domain residual echo estimation erasure unit is subjected to the processing in the wave number domain residual echo estimation erasure unit,
The received signal in the wave number domain used in the wave number domain diffuse residual echo estimation erasing unit is one frame before the received signal in the wave number domain used in the wave number domain residual echo estimation erasing unit.
Echo canceler. The echo canceller according to any one of claims 1 to 4,
Using the received signal in the time domain and the collected sound signal in the time domain, an echo canceling unit that estimates and cancels an echo component included in the collected sound signal in the time domain further includes:
The echo canceller is
A first frequency domain transform unit for transforming the received signal in the time domain into a frequency domain signal;
A first wave number domain converter for converting the received signal in the frequency domain into a signal in the wave number domain;
A multiplier that multiplies the received signal in the wavenumber domain by a filter coefficient in the wavenumber domain to generate an echo replica in the wavenumber domain;
An inverse wave number converter for converting the echo replica in the wave number domain into the echo replica in the frequency domain;
A time domain transforming unit for transforming the echo replica in the frequency domain into the echo replica in the time domain;
A third subtracting unit for subtracting the echo replica in the time domain from the collected sound signal in the time domain to obtain an error signal in the time domain;
A second frequency domain transform unit that transforms the time domain error signal into a frequency domain signal;
A second wavenumber domain converter for converting the error signal in the frequency domain into a signal in the wavenumber domain;
A correction amount calculation unit that calculates a correction amount of the filter coefficient in the wave number domain using the received signal in the wave number domain and the error signal in the wave number domain;
A filter coefficient unit that updates the filter coefficient using the correction amount, and
The collected sound signal used in the wave number domain residual echo estimation cancellation unit or the wave number domain residual echo estimation cancellation unit is processed in the echo cancellation unit, and corresponds to the error signal.
Echo canceler. The echo canceller according to any one of claims 1 to 5,
The wave number domain diffuse residual echo estimation erasure unit is
A residual echo correction unit that corrects each element of the diffuse residual echo vector by multiplying each element of the diffuse residual echo vector by a value based on a value of a lower end of a confidence interval of the estimated value of the input / output transfer characteristic; In addition,
The estimated value of the diffuse residual echo used in the subtraction unit is subjected to processing in the residual echo correction unit.
Echo canceler. E is an echo that eliminates an echo that goes around the microphone via an echo path when P is an integer of 2 or more, P speakers and P microphones are arranged in a common sound field, and a received signal is reproduced from the speakers An erasing method,
Using the signal obtained by converting the collected sound signal collected by the microphone into the wave number domain and the received signal in the wave number domain, the diffusion residual echo included in the collected signal in the wave number domain is estimated, and the wave number domain A wave number domain diffuse residual echo estimation cancellation step for canceling the diffuse residual echo estimated from the collected sound signal,
The wave number domain diffuse residual echo estimation elimination step comprises:
A power spectrum matrix, which is a P × P matrix, is calculated using a received signal vector X, which is a P-dimensional vector having the received signal for each wave number as an element, and its complex conjugate and transpose, and the collected sound signal for each wave number A compressed input / output correlation coefficient calculating step of calculating a cross spectrum matrix that is a P × P matrix using a sound pickup signal vector that is a P-dimensional vector having a component as a component and a complex conjugate and transpose of the received signal vector X; ,
Using the power spectrum matrix and the cross spectrum matrix, compression input to obtain an input / output transfer characteristic matrix which is a P × P matrix having an estimated value of the input / output transfer characteristics of the received signal and the collected sound signal as elements. An output transfer characteristic estimation step;
A diffusion residual echo estimation step of multiplying the reception signal vector X by the input / output transfer characteristic matrix to obtain a diffusion residual echo vector which is a P-dimensional vector having the estimation value of the diffusion residual echo for each wave number as an element;
Subtracting the difference between the collected sound signal in the wave number domain and the estimated value of the diffuse residual echo in the wave number domain,
Echo cancellation method. E is an echo that eliminates an echo that goes around the microphone via an echo path when P is an integer of 2 or more, P speakers and P microphones are arranged in a common sound field, and a received signal is reproduced from the speakers An erasing method,
Using the signal obtained by converting the collected sound signal collected by the microphone into the wave number domain and the received signal in the wave number domain, the diffusion residual echo included in the collected signal in the wave number domain is estimated, and the wave number domain A wave number domain diffuse residual echo estimation cancellation step for canceling the diffuse residual echo estimated from the collected sound signal,
The wave number domain diffuse residual echo estimation elimination step comprises:
P ′ <P, and using a compression matrix W that is a P ′ × P matrix, a received signal vector X that is a P-dimensional vector having the received signal for each wave number as an element is converted into a P′-dimensional compressed vector Z. An input dimension compression step to compress;
A dimension compression matrix updating step for updating the compression matrix W so that a difference between a P-dimensional vector obtained by expanding the compression vector Z by a complex conjugate transpose matrix of the compression matrix W and the received signal vector X is minimized; When,
A power spectrum matrix which is a P ′ × P ′ matrix is calculated using the compression vector Z and its complex conjugate and transpose, and a sound collection signal vector which is a P-dimensional vector having the sound collection signal for each wave number as an element. And a compressed input / output correlation coefficient calculating step of calculating a cross spectrum matrix that is a P × P ′ matrix using the complex conjugate and transpose of the compressed vector Z;
Compression using the power spectrum matrix and the cross spectrum matrix to obtain an input / output transfer characteristic matrix that is a P × P ′ matrix whose elements are estimated values of input / output transfer characteristics of the received signal and the collected sound signal An input / output transfer characteristic estimation step;
A diffusion residual echo estimation step of multiplying the compression vector Z by the input / output transfer characteristic matrix to obtain a diffusion residual echo vector that is a P-dimensional vector whose element is an estimation value of the diffusion residual echo for each wave number;
Subtracting the difference between the collected sound signal in the wave number domain and the estimated value of the diffuse residual echo in the wave number domain,
Echo cancellation method.   A program for causing a computer to function as the echo canceling apparatus according to claim 1.

Images