JP2006217649A - Signal processor - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a signal processor which surely suppresses only the noise, by setting an allowable arrival angle range of a target signal with high precision by a small amount of operation, when the noise is suppressed by performing filter operation processing to signals of a plurality of channels. <P>SOLUTION: Noise is suppressed by an adaptive beam former 11 and the filter operation processing for outptting the signal from a target sound source, inputted from the input direction perpendicular to a straight line connecting the two microphones, is performed to signals of two channels from a microphone array, consisting of two microphones arranged by differing the directions of the directivity of each. <P>COPYRIGHT: (C)2006,JPO&NCIPI

Description

  The present invention relates to a signal processing apparatus that inputs a signal of a plurality of channels and suppresses noise and extracts only a target signal. In particular, the present invention uses a microphone array for voice input of a voice recognition apparatus, a video conference apparatus, and the like. The present invention relates to a signal processing apparatus suitable for a microphone array processing apparatus that suppresses noise and extracts a target voice.

  A microphone array processing apparatus that extracts a target signal from a target sound source by suppressing a noise by using a microphone array composed of a plurality of arranged microphones and processing an output signal of the microphone array is a speech recognition apparatus or Development is underway for the purpose of voice input in video conferencing equipment. Among them, for microphone array processing apparatuses using an adaptive beamformer that can obtain a large noise suppression effect with a small number of microphones, for example, Reference 1: Electronic Information Communication Society edited by “Acoustic System and Digital Processing”, Reference 2: Heykin, As described in “Adaptive Filter Theory (Plentice Hall)”, various methods such as a generalized sidelobe canceller, a frosted beamformer, and a reference signal method are known.

  The adaptive beamformer processing is basically processing for suppressing noise by a filter in which a directional beam having a blind spot in the direction of the noise source is formed. Among them, the generalized sidelobe canceller (GSC) in particular has relatively high performance. Is known for being good. However, GSC has a problem that when the target signal arrives from a direction deviating from the set direction of the target sound source, the target signal is canceled and deteriorated. In order to prevent this, it is necessary to stop the adaptation of GSC when there is a signal arriving from a direction near the set target sound source direction. Therefore, for example, it is disclosed in Reference 3: Julie E. Greenburg et.al .: “Evaluation of anadaptive beamforming method for heating aids”, pp.1662-1676, Jarnul of Acous. Soc. Of Am. 91 (3), 1992. In the hearing aid microphone array processing apparatus, adaptive stop is realized by threshold processing based on the correlation between two microphone output signals.

  However, it is difficult to set an optimum threshold value in the method in which adaptive stop of GSC is performed based on the correlation as described above. That is, the value of the correlation between the two microphone output signals changes depending on the intensity of the target signal and the noise situation even when the target sound source has the same direction and the target signal arrives from a certain direction. The allowable angle range cannot be set with high accuracy. Therefore, for example, if the noise situation changes during voice input, the voice that could be input up to now may not be input temporarily, which is problematic in terms of stability.

  Also, the adaptive beamformer may be distorted without flattening the frequency characteristics when the set direction of the target sound source and the actual direction of arrival of the target signal deviate. If such distortion of frequency characteristics exists, for example, when an adaptive beamformer is used for preprocessing of speech recognition, there is a possibility that the recognition performance is lowered. To solve this problem, the direction of the target sound source to be set is determined by determining the direction of the target sound source set in the adaptive beamformer based on the direction detection process that estimates the sound source direction using a large number of sensors. A method for automatically following the direction of arrival of a signal has been proposed. As another method of direction detection processing, a method called a high resolution method is also known.

  However, in the method of estimating the direction of the target sound source using the sensor, it is difficult to accurately detect the direction of the target sound source when the number of sensors is as small as two, for example. There is a problem that the amount of calculation required for detecting the direction of the target sound source for estimation of the correlation matrix and spatial scanning processing increases.

  As described above, in the microphone array processing apparatus that suppresses noise by adaptive beamformer processing using the conventional GSC, the target signal is canceled when the target signal arrives from a direction deviating from the set target sound source direction. In order to prevent the phenomenon of deterioration, a method for stopping GSC adaptation when there is a signal arriving from a direction near the set target sound source direction by threshold processing based on the correlation between two microphone output signals is as follows. Even if the target signal arrives from a certain direction, the correlation value changes depending on the strength of the target signal and the noise condition, so the allowable arrival angle range of the target signal cannot be set accurately, and noise is generated during voice input. There was a problem in terms of stability, such as temporarily disabling voice input when the situation changed.

  On the other hand, in the method of automatically following the direction of the target sound source by combining the direction detection of the target sound source with many sensors and an adaptive beamformer, it is difficult to detect the direction of the target sound source accurately when the number of sensors is small, Furthermore, the high resolution method has a problem that the amount of calculation required for detecting the direction of the target sound source for estimation of the correlation matrix of the input signal and spatial scanning processing increases.

  The present invention has been made to solve the above-described problems, and in the case where noise is suppressed by performing filter operation processing on a signal of a plurality of channels, it is possible to reliably suppress only noise. To provide an apparatus.

  More specifically, an object of the present invention is to provide a signal processing apparatus that can set an allowable arrival angle range of a target signal with high accuracy with a small amount of calculation.

  It is another object of the present invention to provide a signal processing apparatus that can accurately estimate the direction of arrival of a target signal with a small amount of calculation.

  In order to solve the above-described problem, a first signal processing apparatus according to the present invention includes a microphone array including two microphones arranged with different directivity directions, and two channels from the microphone array. And an adaptive beamformer that performs a filter calculation process for suppressing a noise with respect to a signal and outputting a signal from a target sound source input from an input direction perpendicular to a straight line connecting the two microphones. And

  A second signal processing apparatus according to the present invention suppresses noise with respect to a microphone array composed of two microphones arranged with different directivity directions and two-channel signals from the microphone array. An adaptive beamformer that performs a filter calculation process for outputting a signal from a target sound source that is input from an input direction that coincides with a straight line connecting the two microphones or an input direction that forms an arbitrary angle other than 90 ° to the straight line And a delay unit for matching the phases of the two-channel signals input to the adaptive beamformer when signals arrive from the direction.

  The third signal processing apparatus according to the present invention is a first signal processing unit that performs a filter calculation process for suppressing noise with respect to a signal of a plurality of channels and outputting a signal from a target sound source input from a set input direction. A second adaptive beamformer, and a filter calculation process for suppressing a signal from the target sound source and outputting a signal from a noise source input from a set input direction with respect to the signals of the plurality of channels. The square of the absolute value of the inner product of the adaptive beamformer, the direction vector representing the propagation phase delay of the signals of the plurality of channels, and the frequency components of the filter coefficients generated by the first adaptive beamformer with respect to the frequency Noise source direction estimating means for estimating the direction in which the sensitivity for each direction obtained as a minimum is the direction of the noise source, the direction vector, and the second adaptive beamformer Target sound source direction estimating means for estimating the direction of the target sound source as the direction in which the sensitivity for each direction obtained by adding the square of the absolute value of the inner product of the filter coefficient generated by each frequency component to the frequency is minimized. First control means for setting the direction of the noise source estimated by the noise source direction estimation means as an input direction of the second adaptive beamformer, and the purpose estimated by the target sound source direction estimation means And second control means for setting the direction of the sound source as the input direction of the first adaptive beamformer.

  According to the present invention, it is possible to reliably suppress only noise when performing filter arithmetic processing on a signal of a plurality of channels to suppress noise. Specifically, the arrival direction of a target signal can be accurately determined with a small amount of calculation. Can be estimated.

  That is, according to the present invention, in the microphone array processing apparatus, the microphone array is configured by a plurality of microphones arranged with different directivity directions, and noise is suppressed with respect to signals of a plurality of channels from the microphone array. By performing the filter calculation process for this purpose, it is possible to alleviate the influence of so-called spatial aliasing and to suppress the sensitivity of the direction of the noise source at all frequencies.

  Further, according to the present invention, the second filter calculation means for suppressing the signal from the target sound source is provided separately from the first filter calculation means for suppressing the noise and outputting the signal from the target sound source. By estimating the direction of the target sound source from the directivity of the filter in the second filter calculation means and setting the first filter calculation means based on the estimation result, the correlation matrix estimation, inverse matrix, eigenvalue It is possible to easily follow the signal from the target sound source with a small amount of computation even when computation such as expansion is not required and the number of channels is small.

Hereinafter, embodiments of the present invention will be described with reference to the drawings.
(First embodiment)
In the present embodiment, a signal processing apparatus that controls an adaptive beamformer based on an impulse response between signals of a plurality of channels in order to prevent erroneous cancellation of a target signal in the adaptive beamformer will be described.

  As shown in FIG. 1, the signal processing apparatus according to the present embodiment receives n-channel ch1 to chn signals from input terminals 10-1 to 10n as input, and filters for suppressing noise with respect to these signals. An adaptive beamformer 11 that performs arithmetic processing (adaptive beamformer processing), an impulse response estimation unit 12 that estimates an impulse response between signals of channels ch1 to chn, and an adaptive beamformer 11 according to the estimation result of the impulse response estimation unit 12 And a control unit 13 for controlling the adaptive processing.

  Specifically, this signal processing apparatus is, for example, a microphone array processing apparatus, and n-channel ch1 to chn signals from a microphone array including n microphones (not shown) are input to the input terminals 10-1 to 10n. The Here, for simplicity, a case where the number of channels n is 2 will be described as an example.

  As the processing method inside the adaptive beamformer 11, for example, various methods are known as described in the above-mentioned documents 1 and 2, and a generalized sidelobe canceller (GSC), adaptive noise canceller, and frost type are known. There are beamformers and reference signal methods. The present embodiment can be applied to any adaptive beamformer, but here, a two-channel GSC will be described as an example.

  FIG. 2 shows a configuration of a general Jim-Griffith type GSC in a two-channel GSC as an example of the adaptive beamformer 11. This is a GSC composed of a subtracter 21, an adder 22, a delay unit 23, an adaptive filter 24 and a subtracter 25, as shown in Document 2, for example. Various adaptive filters 24 such as LMS, RLS, and projective LMS can be used, and the filter length La is, for example, La = 50. The delay amount of the delay unit 23 is, for example, La / 2.

  FIG. 3 shows the configuration of the impulse response estimation unit 12. The impulse response estimation unit 12 sequentially estimates an impulse response between signals of a plurality of channels, in this example, an impulse response between signals of the first channel ch1 and the second channel ch2, and includes a delay unit 31, an adaptive filter 32, and It consists of a subtracter 33. In addition, although the cross spectrum method etc. are applicable to estimation of an impulse response, it is desirable to use the adaptive filter 32 as shown in FIG. 3 from the point of computational complexity. The type of the adaptive filter 32 may be anything as described above, but the filter length Lb can be made smaller than the filter length La of the adaptive filter 24 in FIG. 2 in the adaptive beamformer 11, for example, Lb = 10 Is used. Thereby, the amount of calculation of the impulse response estimation part 12 can be made very small.

  In the present embodiment, the impulse response is estimated by the impulse response estimation unit 12 in parallel with the beamformer processing of the adaptive beamformer 11 by using the input signals of the two channels ch1 and ch2, and the estimated impulse response component values are obtained. Based on this, the control unit 13 performs adaptive control of the adaptive beamformer 11. More specifically, this control is control of adaptive stop and adaptive continuation of the adaptive beamformer 11. The adaptive control of the adaptive beamformer 11 can be performed by changing the value of the step size μ when an LMS adaptive filter or a projection LMS adaptive filter is used as the adaptive filter 24.

When an LMS adaptive filter is used as the adaptive filter 24 of the two-channel Jim-Griffith GSC shown in FIG. 2 constituting the adaptive beamformer 11 of FIG. 1, the adaptive filter is updated by setting the time to n as the adaptive filter 24. , The i-th channel input signal x _i (n), and the i-th channel input signal vector X _i (n) = (x _i (n−L−1), x _i (n −L−1),..., X _i (n−L−2)), it is expressed by the following equation.

y (n) = x ₀ (n) + x ₁ (n)
X ′ (n) = X ₁ (n) −X ₀ (n)
e (n) = y (n) -W (n) X '(n)
W (n + 1) = W (n) −μX ′ (n) e (n) (1)
Here, since the adaptation becomes slow if the step size μ is reduced, the value of μ is changed according to the component of the impulse response.

  On the other hand, the impulse response estimation in the impulse response estimation unit 12 can be similarly performed using an LMS adaptive filter as the adaptive filter 32 in FIG. At this time, as shown in FIG. 3, the signal of one channel ch1 of the two channels ch1 and ch2 used for impulse response estimation is delayed by a delay unit 31 by, for example, half Lb / 2 of the filter length Lb to obtain the target signal. The signal of the other channel ch2 is used as a reference signal without being delayed. In this case, if the filter coefficient of the adaptive filter 32 of the impulse response estimation unit 12 is set to Wb, the component of the time delay 0 between the channels ch1 and ch2 is the Lb / 2th value Wb (Lb / 2) of Wb. It becomes.

  In the control unit 13 of FIG. 1, which component value of the impulse response estimated by the impulse response estimation unit 12 is used for adaptive control of the adaptive beamformer 11 depends on the time delay between the channels ch1 and ch2 for the target signal. You may decide based on it.

  For example, as shown in FIG. 4, signals from microphones M1 and M2 are input as signals of two channels ch1 and ch2, and the direction perpendicular to the line connecting both microphones M1 and M2 is the arrival direction of the target signal (target Sound source direction), the value of one component Wb (Lb / 2) corresponding to a time delay of 0 among the components of the impulse response estimated by the impulse response estimator 12 is applied by the adaptive beamformer 11. It is used to control.

  On the other hand, as shown in FIG. 5, the direction perpendicular to the line connecting the two microphones M1 and M2 is set as the direction of the target sound source, and the signal coming from the angle range indicated by ± θ is also the target signal. In the case of allowing, the time delay of the signal arriving from the direction indicated by the arrow a and the time delay of the signal arriving from the direction indicated by the arrow b among the impulse response components estimated by the impulse response estimation unit 12 For the components corresponding to all the time delays between, for example, the maximum value is used for adaptive control of the adaptive beamformer 11.

As shown in FIG. 6, when the distance between the microphones M1 and M2 is d and the sound speed is c, the time delay τ between the signal arrival angle φ and the channels ch1 and ch2 is
τ = dsin (φ) / c (2)
Can be calculated by Thus, the number of the time delay component in the adaptive filter 32 in the impulse response estimation unit 12 is obtained by taking an integer of τ / Ts with Ts as the sampling period. When setting the arrival angle range of the target signal as ± θ in FIG. 5, from τ = dsin (θ) / c obtained by Equation (2), if the integer part of τ / Ts is k,
{Wb (-k + Ld / 2), Wb (-k + Lb / 2 + 1), ... Wb (Lb / 2), Wc (Lc / 2 + 1), ..., Wb (Lb / 2 + k)}
The maximum value among the values is the component value h of the impulse response.

Next, the flow of processing in the signal processing apparatus of the present embodiment described above will be described using the flowchart shown in FIG.
First, an allowable angle range ± θ in the target sound source direction (target signal arrival direction) is set as an initial setting (step S101), and then the impulse response estimation unit 12 inputs the input signal to the adaptive filter 32 to estimate the impulse response. (Step S102). In step S102, the component value h of the impulse response component having a predetermined time delay as described above is obtained, and the component value h is averaged by the following equation to obtain the average value b.
b = h * α + b * (1−α) (3)
Here, α is a forgetting factor, and α = 0.1 is used, for example.

  Next, it is determined whether or not the component value h is a threshold value, that is, whether or not the current component value h is larger than the sum of the threshold value s and the average value b. The signal is output to the adaptive beamformer 11, and if it is smaller, a signal to continue adaptation is output. That is, if h> b + s, the adaptation is stopped, otherwise the adaptation is continued (step S103). For example, s = 0.15 is used as the threshold value s.

  Next, the adaptive beamformer 11 is processed according to, for example, equation (1). At this time, if an adaptive stop signal is output based on the determination result of step S103, the adaptation is stopped by setting the step size μ in the equation (1) to μ = 0 (step S104), and an adaptation continuation signal is received. If it appears, the adaptive processing is performed with μ set to a predetermined value, for example, μ = 0.01 (step S105).

  And the process of step S101-S015 mentioned above is repeated for every sample of an input signal. In order to increase the efficiency of signal input / output, the repetition unit may be a block process for every 256 samples, for example. That is, the adaptive filter processing in step S102 and steps S104 and S105 is performed for 256 samples collectively, and the averaging of the component value h in step S102 and the threshold processing in step S103 is performed once for 256 points. Like that.

  The control of adaptive processing based on the correlation described in Reference 3 focuses only on the correlation related to time delay 0, and thus the control related to the arrival angle of the target signal greatly deviated has become unstable. In the embodiment, the impulse response estimation unit 12 obtains an impulse response between signals of a plurality of channels, and the control unit 13 controls the adaptive processing of the adaptive beamformer 11 based on the impulse response component. It can be set to allow a signal coming from a wide angle range to be a target signal. That is, the allowable arrival angle range of the target signal can be set accurately and over a wide range. Therefore, for example, even when the noise situation changes during the voice input to the microphone, the voice can be input continuously and stably.

(Second Embodiment)
Next, an embodiment in which a directional microphone is used as a microphone used for input of the adaptive microphone array processing apparatus will be described.

  In general, the signal from an omnidirectional microphone or antenna element is used as the input of an adaptive beamformer. For example, an adaptive beamformer using two omnidirectional microphones can generate a frequency band that cannot eliminate noise. .

  FIG. 8 shows a filter in the GSC when a two-channel Jim-Griffith type GSC is used as an adaptive beamformer and is adapted assuming the arrangement of the microphones M1 and M2 and the direction of the noise source shown in FIG. The straight line a is the frequency characteristic of the direction of the target sound source (the direction of arrival of the target signal), and the curve b is the frequency characteristic of the direction of the noise source (the direction of arrival of white noise).

  Curve b shows that the power is reduced by several tens of dB at most frequencies, and that noise removal is working. However, the noise is not removed near the zero frequency and the frequency indicated by fo. ing. This corresponds to a time delay when arrival noise from the set noise source direction reaches each of the microphones M1 and M2 in the case of the arrangement of the microphones M1 and M2 and the direction of the noise source as shown in FIG. This is because the phase coincides with the phase of the time delay of the target signal by an integral multiple of 2π, which is called spatial aliasing. This phenomenon can be avoided when a large number of microphones are used, but it has been difficult to avoid in the case of such a two-channel configuration.

  In order to solve such problems, in this embodiment, directional microphones are used instead of omnidirectional microphones, and these are arranged with different directivity directions to reduce the effect of spatial aliasing. ing. Examples of various arrangements of directivity microphones and directivity patterns in this embodiment are shown in FIG.

  First, in FIG. 10A, two unidirectional microphones are used, and the microphones are arranged so as to be symmetrical by shifting the directivity direction by an angle ± φ with respect to the target direction (the direction of the target sound source). . In order to obtain a sufficient effect of suppressing spatial aliasing, it is desirable to set the value of the shift angle φ in a range of 20 ° ≦ φ ≦ 180 °. The directivity of the microphone need not be limited to the single directivity as shown in FIG. 10A, and may be bidirectional or narrow directivity. If the directivity patterns of the two microphones are the same, a symmetrical arrangement as shown in FIG.

  On the other hand, the directivity patterns of the two microphones are not necessarily the same. For example, as shown in FIG. 10B, one may be omnidirectional and the other may be unidirectional. In this case, it is necessary to adjust the sensitivity so that the sensitivity with respect to the target direction matches between the two microphones. For example, in the case of the arrangement shown in FIG. 10B, an angle at which the sensitivity g1 of the target direction of the unidirectional microphone of the first channel ch1 matches the sensitivity g2 of the omnidirectional microphone of the second channel ch2 is selected. Corrections are made so that the amplitudes coincide with the output signals of both microphones. The signal amplitude can be corrected by multiplying the output signals of both microphones by a coefficient obtained by converting the sensitivity difference between the two microphones into an amplitude.

  Further, the target direction does not need to be perpendicular to the straight line l connecting the two microphones. As shown in FIG. 10C, the straight line l connecting the two microphones coincides with the target direction, or FIG. As shown, the straight line l and the target direction may form an arbitrary angle. In these cases, as shown in FIG. 11, a delay device 14 is arranged in front of the beamformer main body 15 so that the phases of the signals of the channels ch1 and ch2 when the signals arrive from the target direction coincide with each other. The channel ch2 signal is delayed by the delay time d / c. Here, c is a signal propagation speed.

  FIG. 8 shows the frequency characteristics of the filter in the two-channel Jim-Griffith type GSC when the microphone arrangement described above is used. Curves c, d, and e in FIG. 8 are frequency characteristics with respect to the noise arrival direction when the microphones having the same arrangement as in FIG. 9 are made unidirectional and the deviation angles φ = 10 °, 30 °, and 60 °. . From this result, as the deviation angle φ is increased, the spatial aliasing peak that appears in the case of the omnidirectional curve b decreases, and the sensitivity in the direction of the noise source decreases at all frequencies. I understand.

(Third embodiment)
In the present embodiment, even when the target signal arrives from a direction deviating from the direction of the set target sound source, the target signal is not deteriorated by following the direction of the target sound source set in the beam former in that direction. A signal processing apparatus that can output the signal will be described.

  Conventionally, when the direction of arrival of a signal is estimated by, for example, the minimum variance method or the MUSIC method as described in Reference 1, a correlation matrix of an input signal is estimated to obtain an inverse matrix or eigenvector of the correlation matrix, Furthermore, it was necessary to perform a spatial scanning process. However, these processes require a large amount of calculation, and there is a complexity that it is necessary to estimate a correlation matrix in consideration of the ability to follow environmental changes.

  On the other hand, in this embodiment, in order to follow the direction of the target sound source, a second beam former is provided separately from the first beam former that outputs the target signal, and the directivity of the filter in the second beam former is provided. Then, the direction of the target sound source is estimated, and the first beamformer is set based on the estimation result. This eliminates the need for operations such as correlation matrix estimation, inverse matrix, and eigenvalue expansion, and enables tracking of the target signal with a small amount of calculation.

  FIG. 12 shows the configuration of the signal processing apparatus according to the present embodiment. The signals of the channels ch1 and ch2 from the input terminals 40-1 and 40-2 are input to the first and second beamformers 41 and 42. The first beamformer 41 outputs a target signal, and the second beamformer 42 suppresses the target signal and outputs noise.

  The noise source direction estimation unit 43 estimates the direction of the noise source (the arrival direction of the noise signal) from the filter coefficient of the filter in the first beamformer 41, and gives the estimation result to the first control unit 44. The target sound source direction estimation unit 45 estimates the direction of the target sound source (the arrival direction of the target signal) from the filter coefficient of the filter in the second beamformer 42 and gives the estimation result to the second control unit 46. Hereinafter, the direction of the target sound source set in the first and second beam formers 41 and 42 is referred to as an input direction.

  The first control unit 44 controls the second beamformer 42 so that the direction of the noise source estimated by the noise source direction estimation unit 43 is set as the input direction. The second control unit 46 controls the first beamformer 41 so that the direction of the target sound source estimated by the target sound source direction estimation unit 45 is set as the input direction. The reason why the input direction of the second beam former 42 is set to the direction of the noise source is to prevent the second beam former 42 from estimating the direction of the noise source.

  As described above, the first and second beam formers 41 and 42 may be GSC, frost type, or reference signal type. In this case, since a blind spot with low sensitivity is formed in the direction of the noise source in the filter in the first beam former 41 and in the direction of the target sound source in the filter in the second beam former 42, the filter of each filter By examining the directivity from the coefficients, the direction of the noise source and the target sound source can be estimated.

  Generally, the input direction of the beamformer is set by delaying the input signal so that signals from the input direction are in phase with each other as is well known. For this reason, also in the present embodiment, the first and second beam formers 41 and 42 are configured as shown in FIG. When the first and second beamformers 41 and 42 are of the reference signal type, the arrival direction of the reference signal may be matched with the input direction. In the case of the frost type, for example, Reference 4: Kazuaki Takao et. al .: “An adaptive antenna array under directional constraint”, IEEE Teans.on Antenna Propag.Vol.AP-24, No.5, Sep.1976, the incoming signal from any input direction This can also be done by setting linear constraint conditions so as not to attenuate.

  Since the noise source direction estimation unit 43 and the target sound source direction estimation unit 45 estimate the directions of the noise source and the target sound source from the directivities of the filters in the first and second beam formers 41 and 42 as described above, The processing is performed according to the procedure shown in the flowchart of FIG. In addition, although the process in case an input signal is 2 channels is described here as an example, it is not limited to 2 channels.

First, as initial settings, a search range θ _r , a filter length L, an FFT length (FFT point number) N, a channel number M, and the like are set (step S201). In the first beam former 41, in order corresponding to the direction of any noise sources, the search range theta _r a direction of the target sound source as a reference (0 °) and 90 °, the target signal in the second beam former 42 In order to search only within the arrival range, for example, θ _r = 20 °. The search angle range is the range of the input direction ± θ _r . The filter length L and the number of channels M are the same as those of the beam formers 41 and 42, for example, L = 50 and M = 2, respectively, and the FFT length N is a value larger than the filter length L, for example, 64 points in this case. .

  Next, if the beam formers 41 and 42 are GSC, filter conversion is performed (step S203), and if not, FFT is performed (step S204).

In step S203, when the beamformers 41 and 42 are GSC, the filter coefficient is converted into a form equivalent to a transversal beamformer. For example, in the case of a two-channel Jim Griffith type GSC, if the characteristics of the adaptive filter in the GSC are set as wg = (w ₀ , w ₁ , w ₂ ,..., W _L−2 , w _L−1 ), characteristics of the equivalent filter of the channel ch1 _{is, w e1 = (- w 0} , -w 1, -w 2, ..., -w L / 2 + 1, ..., -w L-2, -w L-1), characteristics of the equivalent filter of the second channel ch2 _{is, w e2 = (w 0,} w 1, w 2, ..., w L / 2 -1, ..., w L-2, w L-1) and may be put.

Next, FFT (fast Fourier transform) of the filter coefficient is performed for each channel, and the frequency component W _ei (k) is obtained (step S204). Here, k is a frequency component number, and i is a channel number.

Next, assuming that one direction in the search range is θ, a direction vector S (k, θ) representing a propagation phase delay of each channel related to a signal arriving from the θ direction is generated (step S205). For example, in the case of the microphone arrangement shown in FIG. 6, the direction vector S (k, θ) is given by S (k, θ) = (1, e ^{−jk / Nfsdsin (θ)} ) with reference to the first channel ch1. Become. fs is a sampling frequency.

Then, the frequency components of the filter determined by the _{FFT W e = (W e1 (} k), W e2 (k)) and the direction vector S (k, θ) = ( 1, e -jk / Nfsdsin (θ)) of The absolute value square of the inner product | S · W | ² is obtained (step S206).

The processing in steps S205 to S206 is performed for all frequencies, that is, k = 1 to k = 2 / N, the square sum of the obtained inner products is added for frequency k for each direction θ, and for each direction collected for all bands. Sensitivity
D (θ) = Σ | W (k) S (k, θ) | ² (4)
Is obtained (steps S207 to S208). At this time, the direction is changed by 1 °, for example, and all directions in the search range are examined.

  Next, the direction θmin in which the obtained sensitivity for each direction is minimized is obtained from D (θ), and this is set as the arrival direction of the signal (target signal or noise signal) (step S209). At this time, in order to reduce the influence of the estimation error due to noise, the input direction is updated while the estimated sound source direction is weighted by the output of the beamformer and averaged with the previously estimated sound source direction. Like that. For example, the calculation is performed according to the following equation.

θ ₁ = θ ′ ₁ * g ₁ − (1.0−g ₁ ) * θ ₀
θ ₂ = θ ′ ₂ * g ₂ − (1.0−g ₂ ) * θ _{n
g 1 = 1.0-1 / (f (} p 1 -p th) f (r 1 -r th) +1.0)
_{g 2 = 1.0-1 / (f (} p 2 -p th) f (r 2 -r th) +1.0)
r ₁ = p ₁ / p ₂ , r ₂ = p ₂ / p ₁
Here, g ₁ and g ₂ are weights for updating the input directions of the first and second beam formers 41 and 42, respectively, and p ₁ and p ₂ are blocks of the first and second beam formers 41 and 42, respectively. Unit output power, p _th is a power threshold value, r _th is a power ratio threshold value, θ ′ ₁ and θ ′ ₂ are input directions before updating the first and second beam formers 41 and 42, respectively. , Θ ₁ , θ ₂ are the updated input directions of the first and second beam formers 41, 42, respectively, θ ₀ , θ _n are the newly estimated directions of the target sound source and noise source, f (x) Is a function that gives x if x is positive and 0 if negative.

By such an update, the control is performed so that the update is accelerated when the power of the target sound from the target sound source is large and the power of the noise is small, and the update is delayed otherwise. The values of g ₁ , g ₂ , p _th and r _th are, for example, when the input signal is a 16-bit digital signal and the output power is obtained every 256 points, g ₁ = 0.3, g ₂ = 0.02, Although p _th = 1.0 and r _th = 0.5, it is preferable to determine experimentally according to the situation.

  Next, the overall processing flow of this embodiment including the above-described signal direction estimation processing will be described with reference to the flowchart shown in FIG. Note that the processing of the two-channel array is described here as an example, but the present invention is not limited to two channels and can be applied to multiple channels.

  First, a range Φ allowed as the direction of the target sound source is set as an initial setting, the input direction θ1 of the first beamformer 41 is set to, for example, 0 °, and the input direction θ2 of the second beamformer 42 is set to, for example, 90 °. The search range θr1 of the noise source direction estimation unit 43 is set to 90 °, for example, and the search range θr2 of the target sound source direction estimation unit 45 is set to 20 °, for example (step S301). The allowable range Φ in the target sound source direction is set to the same value as the search range θr of the second beamformer 42, and Φ = θr2 = 20 °. With this setting, a signal that arrives within an angle range of ± Φ is regarded as a target signal. Here, the direction is 0 ° perpendicular to the straight line connecting the two microphones.

  Next, the input direction of the first beamformer 41 is set (step 302). Here, by delaying the signals of the two channels, the signals from the set input direction can equivalently reach the array simultaneously. Therefore, in the first beamformer 41 having the configuration shown in FIG. 11, the delay given to the signal of the first channel ch1 by the delay unit 14 is calculated by τ = dsin (θ1) / c.

  Next, the first beamformer 41 is processed (step S303), and the direction of the noise source is estimated within the search range ± θr1 by the method described above from the obtained filter coefficient (step S304). Let θn be the direction of the estimated noise source.

  Next, it is determined whether or not the direction θn of the noise source estimated in step S304 is in the vicinity (0 ° ± Φ) of the direction of the target sound source (step S305). .

  On the other hand, if the direction θn of the noise source estimated in step S304 is not near the direction of the target sound source, the input direction of the second beamformer 42 is set so that the estimated direction of the noise source is the input direction. (Step S307). That is, the value of θ2 is updated by the above-mentioned averaging. As in step S302, in order to delay the signal of the second channel ch2 so that the signal from the input direction reaches the array at the same time, the second beamformer 42 having the configuration of FIG. The delay given to the first channel ch1 by the device 14 is calculated by τ = dsin (θ2) / c).

  Next, the processing of the second beamformer 42 is performed (step S308), the direction of the target sound source is estimated within the search range ± θr2 (step S309), and the process returns to step S302 again to determine the estimated target sound source. The input direction of the first beamformer 41 is set so that the direction is the input direction. Also at this time, the input direction is updated by the averaging described above.

  Thereafter, the above processing is repeated. The unit of processing may be performed in units of blocks, for example, with 256 points as one block for efficiency. In this case, the processing of the beam formers 41 and 42 uses the filter coefficient after processing one block of data as input for the next step.

  As described above, according to the present embodiment, the direction of the target sound source (the arrival direction of the target signal) is estimated from the filter coefficient of the filter in the second beam former 42 out of the two beam formers 41 and 42. The target signal is accurately extracted with a small amount of computation by suppressing the phenomenon that the target signal is canceled when the target signal arrives from the input direction of each beam former 41, 42 in which the direction of the target sound source is greatly deviated from the setting. It can be performed.

(Fourth embodiment)
In the present embodiment, the signal direction is determined by combining the adaptive control based on the impulse response component described in the first embodiment in the beamformer process using the follow-up to the target signal described in the third embodiment. A signal processing apparatus that improves the accuracy of estimation will be described.

  FIG. 15 shows the configuration of the signal processing apparatus according to the present embodiment. This signal processing apparatus has a configuration in which an impulse response estimation unit 47 is added to the configuration of the third embodiment shown in FIG. The impulse response estimation unit 47 estimates the impulse response between the signals of the channels ch1 and ch2, and gives the impulse response component values, which are the estimation results, to the first and second control units 44 and 46. The first and second control units 44 and 46 control the adaptive processing of the first and second beam formers 41 and 42 based on the component values of the impulse response.

  As described in the first embodiment, the impulse response has a component value corresponding to a time delay between signals of a plurality of channels. Since this time delay corresponds to the arrival direction of the signal, the time delay corresponding to the input direction of the first beamformer 41 is obtained, and when the component value of the impulse response corresponding to this time delay is greater than or equal to a certain value, the first time delay is obtained. The adaptation of the first beamformer 41 is stopped and the second beamformer 42 is adapted, and when the value is below a certain value, the opposite process is performed so that each beamformer 41, 42 receives a signal from its own input direction. Therefore, it is possible to prevent the target signal cancellation phenomenon that is adapted to the above, and it is possible to prevent the deterioration of the signal direction estimation accuracy using the filter coefficient of the adaptive filter. As a result, the overall performance can be improved.

  Next, the processing flow of this embodiment will be described with reference to the flowchart shown in FIG.

  Note that the processing of the two-channel array is described here as an example, but the present invention is not limited to two channels and can be applied to multiple channels.

  First, as in the case of the third embodiment, a range Φ that is allowed as the arrival direction of the target signal is set as an initial setting, the input direction θ1 of the first beamformer 41 is set to 0 °, for example, and the second beamformer 42 is set. Is set to 90 °, for example, 43 search range θr1 of the noise source direction estimation unit is set to 90 °, for example, and search range θr2 of the target sound source direction estimation unit 45 is set to 20 °, for example (step S401). The allowable range Φ in the target sound source direction is set to the same value as the search range θr of the second beamformer 42, and Φ = θr2 = 20 °. With this setting, a signal that arrives within an angle range of ± Φ is regarded as a target signal. Here, the direction is 0 ° perpendicular to the straight line connecting the two microphones.

  Next, the input direction of the first beamformer 41 is set (step 402). Here, by delaying the signals of the two channels, the signals from the set input direction can equivalently reach the array simultaneously. Therefore, in the first beamformer 41 having the configuration shown in FIG. 11, the delay given to the signal of the first channel ch1 by the delay unit 14 is calculated by τ = dsin (θ1) / c.

  Next, similarly, the delay given to the signal of the first channel ch1 is calculated by τ = dsin (θ2) / c, and the input direction of the second beamformer 42 is set (step S403).

  Next, the impulse response estimation unit 47 estimates the impulse response between the signals of the channels ch1 and ch2, and, as described in the first embodiment, the component value h of the determined time delay impulse response and the average value b thereof Is obtained (step S404).

  Next, it is determined whether or not the current component value h is larger than the sum of the threshold value s and the average value b (step S405). If larger, the first beamformer 41 is processed in step S406. If it is smaller, the process proceeds to the process of the first beamformer 41 in step S410. In step S406, adaptation is stopped and the first beamformer 41 is processed. That is, processing is performed with the step size μ of the adaptive filter set to zero. In step S407, normal second beamformer processing is performed without stopping adaptation. That is, the step size μ of the adaptive filter is processed as a predetermined value, for example, 0.01.

  Next, the direction of the target sound source is estimated from the filter coefficient of the second beam former 42 by the method described in the third embodiment (step S408), and the input direction of the first beam former 41 is estimated from the estimated direction. Is set using the averaging described in the third embodiment (step S409), and the process returns to the impulse response estimation process of step S404.

  In step S410, the normal first beamformer 41 is processed without stopping adaptation, and the direction of the noise source is estimated from the obtained filter coefficients by the method described in the first embodiment (step S411).

  Next, it is determined whether or not the estimated noise source direction is in the vicinity (θ1 ± Φ) of the direction of the target sound source (step S412). If it is in the vicinity, the process returns to the impulse response processing in step S404 as it is. If not, the input direction of the second beamformer 42 is set from the estimated direction of the noise source using the averaging described in the third embodiment (step S413), and then the process returns to the impulse response estimation process of step S404. .

  Thereafter, the above processing is repeated. The unit of processing may be performed by block processing with 256 points as one block, for example, for efficiency. In this case, the processing of the first and second beamformers 41 and 42 uses the filter coefficient after processing one block of data as input for the direction estimation processing of the next step.

  As described above, in the present embodiment, the adaptive processing of the two beam formers 41 and 42 is controlled based on the component value of the impulse response between the signals of the channels ch1 and ch2. Deterioration due to signal cancellation and filter deterioration when a signal arrives from the input direction of the signal can be prevented, and the beam formers 41 and 42 follow the target sound source direction and the target signal is extracted with high accuracy. Can do.

  Although the microphone array processing apparatus has been described in the above embodiment, the present invention can be applied to a signal processing apparatus that handles signals from elements other than the microphone array, for example, an antenna array in which a plurality of antenna elements are arranged. The present invention can be basically applied to any signal processing apparatus that suppresses noise (unnecessary signal) by inputting a channel signal and performing filter calculation processing.

The block diagram which shows the structure of the signal processing apparatus which concerns on the 1st Embodiment of this invention. Block diagram showing the configuration of the adaptive beamformer in FIG. The block diagram which shows the structure of the impulse response estimation part in FIG. Diagram showing the relationship between the arrangement of two microphones and the direction of the target sound source Diagram showing the relationship between the arrangement of two microphones, the direction of the target sound source, and the allowable arrival angle range of the target sound source A diagram for explaining the time delay between signals from two microphones A flowchart showing a flow of a series of processing in the embodiment The figure which shows the frequency characteristic of the target sound source direction of a filter, and the noise source direction in 2 channel Jim-Griffith type | mold GSC for demonstrating the 2nd Embodiment of this invention. The figure which shows the relationship between the arrangement | positioning of the microphone corresponding to the frequency characteristic of FIG. 8, and the direction of a target sound source and a noise source The figure which shows the various directivity and arrangement | positioning of two microphones in the embodiment Block diagram showing the configuration of a beamformer with a delay unit inserted on the input side of one channel The block diagram which shows the structure of the signal processing apparatus which concerns on the 3rd Embodiment of this invention. The flowchart which shows the procedure of the estimation process of the direction of a noise source and the target sound source in the embodiment The flowchart which shows the flow of the whole process in the same embodiment The block diagram which shows the structure of the signal processing apparatus which concerns on the 4th Embodiment of this invention. A flowchart showing a flow of processing in the embodiment

Explanation of symbols

DESCRIPTION OF SYMBOLS 10-1-10-n ... Input terminal 11 ... Adaptive beamformer 12 ... Impulse response estimation part 13 ... Control part 14 ... Delayer 15 ... Beamformer main body 21 ... Subtractor 22 ... Adder 23 ... Delayer 24 ... Adaptive filter DESCRIPTION OF SYMBOLS 25 ... Subtractor 31 ... Delay device 32 ... Adaptive filter 33 ... Subtractor 41 ... 1st beam former 42 ... 2nd beam former 43 ... Noise source direction estimation part 44 ... 1st control part 45 ... Target sound source direction estimation Unit 46 ... Second control unit 47 ... Impulse response estimation unit

Claims (3)

A microphone array composed of two microphones arranged with different directivity directions;
An adaptive beam that performs filter calculation processing for suppressing noise with respect to the two-channel signal from the microphone array and outputting a signal from a target sound source input from an input direction perpendicular to a straight line connecting the two microphones A signal processing apparatus comprising: a former. A microphone array composed of two microphones arranged with different directivity directions;
Noise is suppressed with respect to the two-channel signal from the microphone array, and the target sound source is input from an input direction that coincides with a straight line connecting the two microphones or an input direction that forms an arbitrary angle other than 90 ° with respect to the straight line. An adaptive beamformer that performs filter calculation processing to output a signal from
A signal processing apparatus comprising: a delay unit configured to match phases of two-channel signals input to the adaptive beamformer when a signal arrives from the direction. A first adaptive beamformer that suppresses noise with respect to a signal of a plurality of channels and performs a filter calculation process for outputting a signal from a target sound source input from a set input direction;
A second adaptive beamformer that performs a filter calculation process for suppressing a signal from the target sound source with respect to the signals of the plurality of channels and outputting a signal from a noise source that is input from a set input direction;
For each direction obtained by adding the square of the absolute value of the inner product of the direction vector representing the propagation phase delay of the signals of the plurality of channels and each frequency component of the filter coefficient generated by the first adaptive beamformer with respect to the frequency. Noise source direction estimation means for estimating the direction in which the sensitivity of the noise is the minimum as the direction of the noise source,
The direction in which the sensitivity in each direction obtained by adding the square of the absolute value of the inner product of the direction vector and each frequency component of the filter coefficient generated by the second adaptive beamformer with respect to the frequency is minimized A target sound source direction estimating means for estimating the direction of the sound source;
First control means for setting the direction of the noise source estimated by the noise source direction estimation means as an input direction of the second adaptive beamformer;
And a second control unit configured to set the direction of the target sound source estimated by the target sound source direction estimation unit as an input direction of the first adaptive beamformer.

Images