JP4119328B2 - Sound collection method, apparatus thereof, program thereof, and recording medium thereof. - Google Patents

Description

  The present invention relates to a sound collection method and apparatus for TV conferences, audio conferences, telephone calls, remote lectures, and the like.

FIG. 6 is a block diagram of a conventional sound collecting device. This prior art sound collecting device includes microphones 11 _{1 to} 11 _M , adaptive filters 13 B _{1 to} 13 B _M , learning filters 13 A _{1 to} 13 A _M , spatial characteristic filters 18 _{1, 1} to 18 _{J, M,} and signal generator 17 ₁ to 17 _J , delay units 19 _{1 to} 19 _J , sound collection range setting unit 30, virtual target sound source position setting unit 26, spatial characteristic estimation unit 27, adaptation period detection unit 20, adaptation algorithm unit 16, and adders 12 _{1 to} 12 _M , 14A, 14B, 15, 51 _{1 to} 51 _M , 52.

  The prior art sound collecting device in FIG. 6 is a device that suppresses noise and picks up the target sound with high quality, picks up the sound of the sound source within the preset sound collecting range, and falls outside the sound collecting range. Suppresses the noise source. However, the noise source and the target sound are discriminated based on whether the sound source is a stationary signal or an unsteady signal in terms of time. The target sound is assumed to be an unsteady signal such as speech, and the noise source is the air conditioning sound. Is assumed to be a stationary signal. Therefore, non-stationary noise cannot be suppressed.

The signals collected by the microphones 11 _{1 to} 11 _M are respectively filtered by the adaptive filters 13B _{1 to} 13B _M , and then added by the adder 14B and output. The adaptive filters 13B _{1 to} 13B _M are learned so that the sensitivity is high with respect to the sound collection range set by the sound collection range setting unit 30, and the sensitivity is low with respect to the noise source position outside the sound collection range. Therefore, the output of the adder 14B is a high-quality sound with a high target sound-to-noise ratio (SN ratio). However, since the sound collection device of the prior art always restricts sensitivity to the entire sound collection range, there is a problem that the noise suppression performance becomes lower as the sound collection range becomes wider.

Next, learning of the adaptive filters 13B _{1 to} 13B _M will be specifically described. Learning is performed using the noise actually collected, the virtual collected sound signal synthesized using the virtual target sound source, and the learning filter. The reason why the virtual target sound source is used in this way is that when the actual target sound source is observed, it is always observed as a signal in which noise is mixed, and therefore processing that distinguishes the target sound from the noise cannot be performed.

First, a portion for synthesizing a virtual sound pickup signal using a virtual target sound source will be described. The sound collection range setting unit 30 sets a sound collection range (sound source movement range, sound source position measurement error range, etc.), and the virtual target sound source position setting unit 26 uniformly sets the virtual target sound source position within the set range. Is provided. The interval between virtual target sound source positions must be sufficiently narrow so that when the sound source moves from one virtual target sound source position to an adjacent virtual target sound source position, the relative delay time variation between microphones becomes smaller than the sampling period. Set the interval. The spatial characteristic estimation unit 27 estimates an impulse response from the set virtual target sound source position to the microphone position _, and sets the coefficients of the spatial characteristic filters 18 _{1, 1} to 18 _{J, M.} The uncorrelated and stationary signals generated by the signal generators 17 _{1 to} 17 _J are filtered by the spatial characteristic filters 18 _{1, 1} to 18 _{J, M} and added by adders 51 _{1 to} 51 _M for each microphone. . Thus, by filtering the signal with the spatial characteristic filters 18 ₁ , ₁ to 18 _{J, M} , the target sound collection signal can be virtually synthesized.

Then, actually adding the collected noise signal by the adder 12 ₁ to 12 _M and virtually synthesized target sound collected sound signal, and filtering the learning filter 13A ₁ ~13A _M this, adding Adder 14A adds. The output of the adder 14A is an output of a virtually collected sound signal. If the noise component of the output is small and the degradation of the virtual target sound component is small, it means that the sound is collected with high quality. Therefore, the adder 15 subtracts the original sound of the virtual target sound from the output. The learning filters 13A _{1 to} 13A _M are updated using the output of 15 as an error signal. However, subtraction to enable non-causal portion of the learning filter 13A ₁ ~13A _M, after adding a delay to the original sound of the virtual target sound by the delaying unit 19 ₁ ~ 19 _J, those added by the adder 52 is doing.

Adaptive algorithm unit 16, the input signal of the adder 15 error signal output to the learning filter 13A ₁ ~13A _M, the update vector of the learning filter 13A ₁ ~13A _M as mean square error of the error signal is minimized Ask. The same filter coefficients as the learning filters 13A _{1 to} 13A _M are set in the adaptive filters 13B _{1 to} 13B _M , the sound of the target sound source within the set sound collection range is collected, and the noise is suppressed. Further, when the actual target sound is included in the collected sound signals of the microphones 11 _{1 to} 11 _M , learning is performed so as to reduce the sensitivity with respect to the actual target sound source, so that there is an actual target sound. Need to stop updating the filter. The adaptation period detection unit 20 detects the presence of the actual target sound by monitoring the power of the signals collected by the microphones 11 _{1 to} 11 _M , and stops the adaptation operation.

Next, the adaptive algorithm unit 16 will be described in detail. Examples of adaptive algorithms include LMS algorithm, NLMS algorithm, and projection algorithm. In this specification, the NLMS method is taken as an example, and a filter convergence solution and a correction formula are derived below. First, symbols used in mathematical expressions will be described. The time sampled by the sampling period is n, the number of microphones is M, the number of virtual target sound sources is J, the signal picked up by the i-th microphone 11 _i at time n is x _i (n), and L samples are extracted. What the matrix represents

And The output signal of the j-th signal generator 17 _j is represented by v _j (n), and the spatial characteristic filter for the j-th signal generator 17 _j and the i-th microphone 11 _i is represented by g _i , _j (n). Filter output is u _{i ,, j} (n) = g _{i, j} (n) * v _j (n)

And However, * represents a convolution operation. The learning filters 13A _{1 to} 13A _M and the adaptive filters 13B _{1 to} 13B _M are L-tap FIR filters, and the filter coefficients

As a matrix. h _i (n −l−1) represents the filter coefficient of the l-th tap of the filter for the i-th microphone at time n, and the same filter coefficient is used for the learning filters 13A _{1 to} 13A _M and the adaptive filters 13B _{1 to} 13B _M. Is used. The output of the adder 14A is y '(n), the output of the adder 14B is y (n), the output of the adder 15 is error e (n), and the delay amounts in the delay units 19 _{1 to} 19 _J are all equal. and it was expressed in tau ₀ (typically, tau ₀ is the length of half of the tap length of the learning filter 13A ₁ ~13A _M).

  First, the mean square of the error e (n) that is the output of the adder 15 is obtained. The filter that minimizes the mean square error is the optimum filter.

However, ￣ means time average. Since the virtual target signal v _j (n) is uncorrelated with each other and the virtual target signal and noise are uncorrelated, Equation (1) is transformed into Equation (2).

Adaptive filter

If L is a L-tap FIR filter and equation (2) is expressed as a vector, equation (3) is obtained.

Since the filter that minimizes Equation (3) is the optimal filter, Equation (3)

To obtain a local minimum point.

Equation (4)

Solving for, the optimal filter that minimizes Equation (3)

Is required.

  There are adaptive algorithms such as an LMS algorithm, an NLMS algorithm, and a projection algorithm as a method for obtaining the optimum filter of Expression (5). In this specification, the NLMS algorithm will be described as an example, and the correction formula is expressed by Formula (6).

However,

Is represented by equation (7).

Up to this point, it has been shown that the optimum filter of equation (5) is obtained using the modified equation of equation (6).
JP-A-14-062895

  However, the above-described conventional sound collection methods have a problem that non-stationary noise signals (for example, speaker voices that are not desired to be collected) cannot be suppressed, and noise suppression performance is improved by widening the sound collection range. There is a problem of lowering.

  An object of the present invention is to provide a sound collection method, apparatus, program, and recording medium that realizes sound collection with suppressed non-stationary noise signals and achieves high suppression performance regardless of the range of the sound collection range. It is.

  To achieve the above object, the sound collection method of the present invention detects a speaker position from a sound collection range setting stage for setting a sound collection range and a sound reception signal received by each of a plurality of sound collection means. The speaker position detection stage, and if the detected speaker position is within the sound collection range, the speaker voice is collected, and if the detected speaker position is outside the sound collection range, the speaker voice is suppressed. A filter coefficient setting stage for setting a filter coefficient using the received signal; a filter stage for filtering each received sound signal received by each of the plurality of sound collecting means with the filter coefficient; and an output signal of each filter stage There is an addition stage for adding.

As a result, it is possible to collect only the sound within the set sound collection range and suppress other sounds.
According to the embodiment of the present invention, the filter coefficient setting stage includes an FFT stage for converting the received signal received by each of the plurality of sound collecting means into the frequency domain, and each of the output signals of the FFT stage as frequency components. A covariance matrix calculation stage to obtain a covariance matrix by multiplying each, a covariance matrix storage stage for averaging the covariance matrix for each detected speaker position, and storing, and a stored covariance matrix And a filter coefficient calculation step of calculating a filter coefficient using the detected speaker position and the sound collection range.
According to the embodiment of the present invention, the filter coefficient setting stage includes an FFT stage for converting the received signal received by each of the plurality of sound collecting means into the frequency domain, and each of the output signals of the FFT stage as frequency components. A covariance matrix calculation stage to obtain a covariance matrix, a covariance matrix storage stage for averaging the covariance matrix for each detected speaker position, and a stored covariance matrix A whitening stage that multiplies the stored covariance matrix by a gain that smoothes the frequency characteristic of the diagonal component with the highest power or the added value of the diagonal component of the stored covariance matrix, and white And a filter coefficient calculation step of calculating a filter coefficient using the detected covariance matrix and the detected speaker position and the sound collection range.

  In order to solve the above-described problems, another sound collection method of the present invention includes a sound collection range / volume range setting stage for setting a sound collection range and a sound volume range, and a sound reception received by each of a plurality of sound collection means. A speaker position detecting stage for detecting a speaker position from a sound signal; a speaker volume estimating stage for estimating a speaker volume from a received sound signal received by each of a plurality of sound collecting means; and a detected speaker If the position is within the sound collection range and the estimated speaker volume is within the volume range, sound is collected; otherwise, the speaker sound suppression is performed. A filter coefficient setting step for setting a filter coefficient using the received sound signal; a filter step for filtering the received sound signal received by each of the plurality of sound collecting means with the filter coefficient; and An adding stage for adding the output signals;

  By adding the condition of the volume range in addition to the condition of the sound collection range, it is possible to suppress only the unnecessary speech of the speaker away from the sound collection means.

  According to the embodiment of the present invention, the filter system number setting stage includes an FFT stage for converting the received signal received by each of the plurality of sound collecting means into a frequency domain, and each of the output signals of the FFT stage as frequency components. A covariance matrix calculation stage to obtain a covariance matrix by multiplying each, a covariance matrix storage stage for averaging the covariance matrix for each detected speaker position, and storing, and a stored covariance matrix And a filter coefficient calculation step of calculating a fill coefficient using the detected speaker position and the sound collection range, and the estimated speaker volume and the sound volume range.

  According to the embodiment of the present invention, the filter coefficient setting stage includes an FFT stage for converting the received signal received by each of the plurality of sound collection means into the frequency domain, and each of the output signals of the FFT stage as frequency components. A covariance matrix calculation stage to obtain a covariance matrix, a covariance matrix storage stage for averaging the covariance matrix for each detected speaker position, and a stored covariance matrix A whitening stage that multiplies the stored covariance matrix by a gain that smoothes the frequency characteristic of the diagonal component with the highest power or the added value of the diagonal component of the stored covariance matrix, and white A filter coefficient calculating step of calculating a filter coefficient using the normalized covariance matrix, the detected speaker position and the sound collection range, and the estimated speaker volume and the sound volume range.

  By whitening the covariance matrix, a filter that does not depend on the frequency characteristics of the sound source can be obtained. Thereby, even if the frequency characteristic of the sound source changes, there is no change in the filter, and it is possible to prevent a change in sound due to the processing of the present invention.

  The present invention sets the sound collection range as described above, collects the sound when the detected speaker position is within the sound collection range, and suppresses the sound outside the range. Since the sound outside the range is suppressed regardless of whether it is stationary or non-stationary, it is possible to suppress the voice that is not desired to be collected. In addition, since only the sensitivity to the sound source position where the sound is actually generated is controlled, the suppression performance does not deteriorate depending on the width of the sound collection range.

[First Embodiment]
FIG. 1 is a block diagram of a sound collecting apparatus according to a first embodiment of the present invention.

The sound collection device according to the first embodiment includes microphones 11 _{1 to} 11 _M , a speaker position detection unit 23, a sound collection range setting unit 25, a filter coefficient setting unit 24, filter units 21 _{1 to} 21 _M, and an adder 22. Is done.

The sound collection range setting unit 25 sets a sound collection range. The sound collection range is set by the user with a button or a remote control, or is fixedly set in advance. Speaker position detection unit 23 includes a sound receiving the signal by the microphone 11 ₁ to 11 _M, detects the speaker position from the position of the microphone 11 ₁ to 11 _M. The filter coefficient setting unit 24 calculates a filter coefficient so that sound is collected if the detected speaker position is within the sound collection range set by the sound collection range setting unit 24, and is suppressed if the speaker position is out of the range. . The calculated filter coefficient is copied to the filter units 21 _{1 to} 21 _M. The filter units 21 _{1 to} 21 _M respectively filter the signals received by the microphones 11 _{1 to} 11 _M. The output signals of the filter units 21 _{1 to} 21 _M are added by the adder 22 to become an output signal. As described above, it is possible to obtain an output signal that collects only the sound within the sound collection range and suppresses unnecessary sound outside the sound collection range.

  The speaker position detection unit 23 and the filter coefficient setting unit 24 will be described in detail below.

  The speaker position detection unit 23 is realized by the following method, for example.

The covariance matrix is calculated from the microphones 11 _{1 to} 11 _M, and the sound power for each scanning position is estimated by multiplying the covariance matrix by the steering vector set for each scanning position. The scanning position having the maximum power is detected as the speaker position from the estimated voice power for each scanning position.

  This will be described below using mathematical formulas.

First, let x _i (t) be the signal received by the i-th microphone 11 _i , and let X _i (ω) be the converted signal in the frequency domain, and input signal vector

Is defined by equation (8).

Where ^T is the transpose of the matrix.

Next, the covariance matrix

Is represented by equation (9).

Where ^H is the conjugate transpose of the matrix.

  Next, a steering vector used for speech power estimation will be described. The steering vector is set so that the sound coming from the scanning position has the same phase. By using such a steering vector, only a signal having the same phase (sound generated at the scanning position) is emphasized, and a sharp directivity is formed at the scanning position.

First, in the case of the scanning position (x, y, z), the delay amount d _i (x, y, z) given to the i-th microphone 11 _i is the sound emitted from the scanning position (x, y, z). From the scanning position (x, y, z), the i-th microphone position (x _i , y _i, z _i ), and the sound velocity c, using the equations (10) and (11). Desired.

However, D is a fixed delay amount and is given as a constant in advance.

  An expression obtained by converting Expression (10) into the frequency domain is Expression (12). A vector obtained by converting the expression into a vector is a steering vector, which is Expression (13).

This steering vector

Multiplied by the covariance matrix and integrated over the frequency, the estimated audio power corresponding to each scan position

Is required. This is expressed by equation (14).

Steering vector

Shows the estimated sound power because only the sound generated at the scanning position (x, y, z) is emphasized with the same phase.

Takes a large value only when there is a sound source at the scanning position. Therefore, the estimated voice power

If the scanning position (x _m , y _m , z _m ) with the maximum power is detected, the speaker position can be estimated.

  Next, the filter coefficient setting unit 24 will be described in detail.

  The filter coefficient setting unit 24 determines whether or not the speaker position detected by the speaker position detection unit 23 is within the sound collection range. When the sound is within the sound collection range, the sound is to be collected, and the others are the suppression targets.

The filter that collects only the sound within the sound collection range and suppresses other sounds is the input signal vector to be collected.

Filter

The signal that has been filtered and added in the mixing belt is the input signal to be collected.

The input signal vector to be suppressed.

Filter

It is only necessary that the signal filtered and added at 0 is 0. Therefore, the filter is optimal when the following expressions (15), (16), and (17) are satisfied.

Filter equations (15)-(17) with least squares solution

If it solves about, it will become a formula (18).

However, C _Sj and C _Nk are the weights of speaker voice collection and suppression weights, respectively. If C _Nk is increased, the amount of suppression of unnecessary voices increases, and if C _Sj is increased, the voices to be collected are collected. Deterioration is reduced.

In order to obtain the filter coefficient from equation (18), it is necessary to obtain the covariance matrix of the input signal for each speaker position. In the present invention, the covariance matrix obtained by equation (9)

Save the time average for each speaker. At this time, the covariance matrix for the target speaker position is

And the covariance matrix for the speaker position to be suppressed is

And

  The filter coefficient can be obtained from the covariance matrix obtained as described above by Equation (18).

  As described above, in the present embodiment, it is possible to collect only the sound within the set sound collection range and suppress other sounds.

  FIG. 5 is a diagram illustrating an example of use of the present invention. It is assumed that the sound collection device using the present invention is placed on a table and a speaker is around it. The device has a mute button for each range. By pressing the mute button, only the sound in the range corresponding to the button can be muted (no sound is collected). In the present invention, a range in which sound is not collected can be set regardless of the steadiness or non-stationarity of the sound, and thus such a utilization method is also possible.

[Second Embodiment]
FIG. 2 is a block diagram of a sound collecting apparatus according to the second embodiment of the present invention.

  The sound collection device of the second embodiment is an example in which a sound collection range / volume range setting unit 31 and a speaker volume estimation unit 32 are added to the sound collection device of the first embodiment.

The sound collection range / volume range setting unit 31 sets a sound collection range and a volume range. The setting is performed by the user with a button or a remote controller, or given in advance. The speaker volume estimation unit 32 estimates the power of the audio signal from the signals received by the microphones 11 _{1 to} 11 _M. If the speaker position detected by the speaker position detection unit 23 is within the sound collection range and the estimated speaker volume is within the volume range, the sound is collected, otherwise the speaker sound is suppressed. To do. Thus, for example, only the sound pickup great voice nearby have received sound power to the microphone 11 ₁ to 11 _M, it is possible to suppress the voice of the speaker away from the microphone 11 ₁ to 11 _M.

The speaker volume estimation method will be described below. Speaker volume

Is the input signal vector

Mixing vector

Is obtained by averaging in the frequency low band W, and is obtained by the equation (19).

  From equation (19), it can be seen that the speaker volume can be estimated from the covariance matrix. Therefore, the covariance matrix can be obtained from equation (9), and the speaker volume can be obtained from equation (19).

In the second embodiment, in addition to the sound collection range condition of the first embodiment, by adding the sound volume range condition, it is also possible to suppress only the unnecessary speech of the speaker away from the microphones 11 _{1 to} 11 _M. It becomes possible.

  Since other parts are the same as those in the second embodiment, description thereof is omitted.

[Third Embodiment]
FIG. 3 is a block diagram of a sound collecting apparatus according to the third embodiment of the present invention. The sound collection device according to the third embodiment is the same as the sound collection device according to the first embodiment or the second embodiment, except that the filter coefficient setting unit 12 includes the FFT units 41 _{1 to} 41 _M and the covariance matrix calculation unit 42. This is an example realized by the dispersion matrix storage unit 43 and the filter coefficient calculation unit 44.

The FFT units 41 _{1 to} 41 _M respectively convert the signals received by the microphones 11 _{1 to} 11 _M into the frequency domain. The covariance matrix calculation unit 42 multiplies the FFT output signal between the channels, and obtains a covariance matrix by Expression (9). The covariance matrix storage unit 43 averages and stores the covariance matrix for each speaker position. The filter coefficient calculation unit 44 calculates the filter coefficient according to the equation (18).

  Since other parts are the same as those in the first embodiment or the second embodiment, description thereof will be omitted.

[Fourth Embodiment]
FIG. 4 is a block diagram of a speaker position detecting apparatus according to a fourth embodiment of the present invention. In the sound collection device of the fourth embodiment, in the sound collection device of the first embodiment or the second embodiment, the covariance matrix calculation unit 12 includes FFT units 41 _{1 to} 41 _M , a covariance matrix calculation unit 42, and the like. This is an example realized by the covariance matrix storage unit 43, the whitening unit 45, and the filter coefficient calculation unit 44.

The FFT units 41 _{1 to} 41 _M , the covariance matrix calculation unit 42, the covariance matrix storage unit 43, and the filter coefficient calculation unit 44 perform the same processing as in the third embodiment, and thus description thereof is omitted.

The whitening unit 45 is a covariance matrix.

Is whitened (to a flat frequency characteristic) in the frequency domain. Whitening is the most powerful of the diagonal components of the covariance matrix

Whitening gain to smooth

Or a whitening gain that smoothes the mean power of the diagonal components of the covariance matrix

This is done by multiplying These are represented by the equations (20) and (21), respectively.

  However, β is a coefficient for adjusting the degree of whitening. When it is 1, it becomes complete whitening, and when it becomes 0, whitening is not performed.

  In the fourth embodiment, it is possible to obtain a filter that does not depend on the frequency characteristics of the sound source by whitening the covariance matrix. Thereby, even if the frequency characteristic of the sound source changes, there is no change in the filter, and a change in timbre due to the processing of the present invention can be prevented.

  Since other parts are the same as those in the first embodiment or the second embodiment, description thereof will be omitted.

  The sound collecting method of the present invention is not only realized by dedicated hardware, but a program for realizing the function is recorded on a computer-readable recording medium, and the program recorded on the recording medium is recorded. May be read by a computer system and executed. The computer-readable recording medium refers to a recording medium such as a floppy disk, a magneto-optical disk, a CD-ROM, or a storage device such as a hard disk device built in the computer system. Furthermore, a computer-readable recording medium is a server that dynamically holds a program (transmission medium or transmission wave) for a short period of time, as in the case of transmitting a program via the Internet, and a server in that case. Some of them hold programs for a certain period of time, such as volatile memory inside computer systems.

It is a block diagram which shows the speaker position detection apparatus of the 1st Embodiment of this invention. It is a block diagram which shows the speaker position detection apparatus of the 2nd Embodiment of this invention. It is a block diagram which shows the speaker position detection apparatus of the 3rd Embodiment of this invention. It is a block diagram which shows the speaker position detection apparatus of the 4th Embodiment of this invention. It is a figure explaining the usage example of this invention. It is a block diagram which shows the example of the conventional speaker position detection apparatus.

Explanation of symbols

11 _{1 to} 11 _M microphones 21 _{1 to} 21 _M filter unit 22 adder 23 speaker position detection unit 24 filter coefficient setting unit 25 sound collection range setting unit 31 sound collection range / volume range setting unit 32 speaker volume estimation unit 41 ₁ ˜41 _M FFT unit 42 covariance matrix calculation unit 43 covariance matrix storage unit 44 filter coefficient calculation unit 45 whitening unit 12 _{1 to} 12 _M adder 13A _{1 to} 13A _M learning filter 13B _{1 to} 13B _M adaptive filter 14A adder 14B adder 15 adder 16 adaptive algorithm unit 17 _{1 to} 17 _J signal generator 18 _{1, 1} to 18 _{J, M} spatial characteristic filter 19 _{1 to} 19 _J delay unit 20 adaptive period detector 51 _{1 to} 51 _M adder 52 Adder 26 Virtual sound source position setting unit 27 Spatial characteristic estimation unit 30 Sound collection range setting unit

Claims (14)

A sound collection method,
A sound collection range setting stage for setting the sound collection range;
A speaker position detection stage for detecting a speaker position from a received sound signal received by each of a plurality of sound pickup means;
When the detected speaker position is within the sound collection range, the voice signal is collected, and when the detected speaker position is outside the sound collection range, the sound reception signal is A filter coefficient setting stage to set the filter coefficient using,
A filter stage for filtering each received signal received by each of the plurality of sound collecting means with the filter coefficient;
A sound collection method comprising an addition step of adding the output signals of the filter step. A sound collection method,
Sound collection range / volume range setting stage to set the sound collection range and volume range,
A speaker position detection stage for detecting a speaker position from a received sound signal received by each of a plurality of sound pickup means;
A speaker volume estimation stage for estimating a speaker volume from a received signal received by each of a plurality of sound collection means;
If the detected speaker position is within the sound collection range and the estimated speaker volume is within the volume range, sound is collected; otherwise, the speaker sound is suppressed. A filter coefficient setting step for setting a filter coefficient using the received sound signal;
A filter stage for filtering each received signal received by each of the plurality of sound collecting means with the filter coefficient;
And a summing step of summing the output signals of the filter step. The filter coefficient setting step includes:
An FFT stage for converting a received sound signal received by each of the plurality of sound collecting means into a frequency domain;
A covariance matrix calculating step of multiplying each of the output signals of the FFT step for each frequency component to obtain a covariance matrix;
A covariance matrix storage step of averaging and storing the covariance matrix for each detected speaker position;
And a filter coefficient calculating step of calculating a filter coefficient using the stored covariance matrix, the detected speaker position and the sound collection range, and the estimated speaker volume and the volume range. Item 3. The sound collection method according to Item 2 . The filter coefficient setting step includes:
An FFT stage for converting a received sound signal received by each of the plurality of sound collecting means into a frequency domain;
A covariance matrix calculating step of multiplying each of the output signals of the FFT step for each frequency component to obtain a covariance sequence;
A covariance matrix storage step of averaging and storing the covariance matrix for each detected speaker position;
The stored covariance is a gain for smoothing the frequency characteristic of the diagonal component of the stored covariance matrix having the highest power or the sum of the diagonal components of the stored covariance matrix. A whitening stage to multiply the matrix;
And a filter coefficient calculation step of calculating a filter coefficient using the whitened covariance matrix, the detected speaker position and the sound collection range, and the estimated speaker volume and the volume range. Item 3. The sound collection method according to Item 2 .   The filter coefficient setting step includes:
  An FFT stage for converting a received sound signal received by each of the plurality of sound collecting means into a frequency domain;
  A covariance matrix calculating step of multiplying each of the output signals of the FFT step for each frequency component to obtain a covariance matrix;
  A covariance matrix storage step of averaging and storing the covariance matrix for each detected speaker position;
  The sound collection method according to claim 1, comprising: a filter coefficient calculation step of calculating a filter coefficient using the stored covariance matrix, the detected speaker position, and the sound collection range.   The filter coefficient setting step includes:
  An FFT stage for converting a received sound signal received by each of the plurality of sound collecting means into a frequency domain;
  A covariance matrix calculating step of multiplying each of the output signals of the FFT step for each frequency component to obtain a covariance sequence;
  A covariance matrix storage step of averaging and storing the covariance matrix for each detected speaker position;
  The stored covariance is a gain for smoothing the frequency characteristic of the diagonal component of the stored covariance matrix having the highest power or the added value of the diagonal components of the stored covariance matrix. A whitening stage to multiply the matrix;
  The sound collection method according to claim 1, further comprising a filter coefficient calculation step of calculating a filter coefficient using the whitened covariance matrix and the detected speaker position and the sound collection range. A sound collecting device,
A sound collection range setting means for setting a sound collection range;
Speaker position detecting means for detecting a speaker position from a received sound signal received by each of a plurality of sound collecting means;
When the detected speaker position is within the sound collection range, the voice signal is collected, and when the detected speaker position is outside the sound collection range, the received signal is used under the condition of suppressing the speaker voice. Filter coefficient setting means for setting the filter coefficient
Filter means for filtering received sound signals received by each of the plurality of sound collecting means, respectively, with the filter coefficients;
A sound collection device having addition means for adding the output signals of the filter means. A sound collecting device,
A sound collection range / volume range setting means for setting a sound collection range and a volume range;
Speaker position detecting means for detecting a speaker position from a received sound signal received by each of a plurality of sound collecting means;
Speaker volume estimation means for estimating speaker volume from a received sound signal received by each of a plurality of sound collection means;
If the detected speaker position is within the sound collection range and the estimated speaker volume is within the volume range, sound is collected; otherwise, the speaker sound is suppressed. Filter coefficient setting means for setting a filter coefficient using the received sound signal;
Filter means for filtering received sound signals received by each of the plurality of sound collecting means, respectively, with the filter coefficients;
A sound collection device having addition means for adding the output signals of the filter means. The filter coefficient setting means includes
FFT means for converting the received sound signal received by each of the plurality of sound collecting means into a frequency domain;
A covariance matrix calculating means for multiplying each output signal of the FFT means for each frequency component to obtain a covariance matrix;
Covariance matrix storage means for averaging the covariance matrix for each detected speaker position and storing the covariance matrix;
And a filter coefficient calculation means for calculating a fill coefficient using the stored covariance matrix, the detected speaker position and the sound collection range, and the estimated speaker volume and the volume range. 8. The sound collecting device according to 8 . The filter coefficient setting means includes
FFT means for converting the received sound signal received by each of the plurality of sound collecting means into a frequency domain;
A covariance matrix calculating means for multiplying each output signal of the FFT means for each frequency component to obtain a covariance matrix;
Covariance matrix storage means for averaging the covariance matrix for each detected speaker position and storing the covariance matrix;
The stored covariance is a gain for smoothing the frequency characteristic of the diagonal component of the stored covariance matrix having the highest power or the sum of the diagonal components of the stored covariance matrix. Whitening means for multiplying the matrix;
Filter coefficient calculating means for calculating a filter coefficient using the whitened covariance matrix, the detected speaker position and the sound collection range, and the estimated speaker volume and the sound volume range; The sound collection device according to claim 8 .   The filter coefficient setting means includes
  FFT means for converting the received sound signal received by each of the plurality of sound collecting means into a frequency domain;
  A covariance matrix calculating means for multiplying each output signal of the FFT means for each frequency component to obtain a covariance matrix;
  Covariance matrix storage means for averaging the covariance matrix for each detected speaker position and storing the covariance matrix;
  The sound collection device according to claim 7, further comprising: a filter coefficient calculation unit that calculates a filter coefficient using the stored covariance matrix, the detected speaker position, and the sound collection range.   The filter coefficient setting means includes
  FFT means for converting the received sound signal received by each of the plurality of sound collecting means into a frequency domain;
  A covariance matrix calculating means for multiplying each output signal of the FFT means for each frequency component to obtain a covariance matrix;
  Covariance matrix storage means for averaging the covariance matrix for each detected speaker position and storing the covariance matrix;
  The stored covariance is a gain for smoothing the frequency characteristic of the diagonal component of the stored covariance matrix having the highest power or the added value of the diagonal components of the stored covariance matrix. Whitening means for multiplying the matrix;
  The sound collection device according to claim 7, further comprising filter coefficient calculation means for calculating a filter coefficient using the whitened covariance matrix, the detected speaker position, and the sound collection range. A sound collection program for causing a computer to execute the speaker position detection method according to claim 1 . A computer-readable recording medium on which the sound collecting program according to claim 13 is recorded.

Images