WO2023214571A1 - Beamforming method and beamforming system - Google Patents

Abstract

In a binaural beamformer 1 to which an algorithm for MVDR-IC is applied, a parameter for controlling a trade-off between IC preservation of a noise component caused by this algorithm and noise suppression can be adjusted from outside, and thus it is possible to adjust the parameter by a user himself or herself or automatically depending on a hearing environment, and achieve appropriate beamforming.

Description

Beamforming method, beamforming system

The present invention relates to beamforming, particularly to a binaural beamforming method using minimum variance distortionless response (hereinafter referred to as "MVDR"), and to devices and systems applying the method. .

Binaural beamforming using MVDR is an algorithm that is guaranteed to preserve the desired audio spatial information. However, it is known that when this algorithm is used, the spatial information of the noise is distorted, and the noise is perceived as coming from the same direction as the desired voice (see, for example, Non-Patent Document 1).

In addition, interaural coherence (hereinafter referred to as "IC"), binaural cues such as interaural level difference and interaural time difference determine the localization width and diffusivity of the sound source. , is known to play a major role in spatial perception. This is because spatial separation of desired speech and noise helps improve the speech reception threshold (hereinafter referred to as "SRT") (for example, Non-Patent Documents 2-4 ).

Under the above-mentioned background, an MVDR-IC holding an IC has been proposed. By using the MVDR-IC, it is possible to process a desired audio binaural cue and a diffuse noise IC while maintaining it. However, the MVDR-IC algorithm has a problem in that there is a trade-off between noise suppression and IC preservation.

Therefore, the present invention aims to provide a technology that realizes appropriate beamforming.

In order to solve the above problems, the present invention employs the following beamforming method and a beamforming system (beamforming device) to which this method is applied. Note that the following words in parentheses are merely examples, and the present invention is not limited thereto.

That is, the beamforming method of the first aspect of the present invention is a beamforming method in which input signals corresponding to sounds input to a plurality of microphones are beamformed using MVDR, and a predetermined design is made. The method includes a filter updating step in which coefficients are calculated based on the result of passing an input signal through an MVDR filter and the FIR filter is switched using the coefficients, and a convolution step in which the input signal is convolved with the FIR filter.

Preferably, in the beamforming method of the first aspect, the MVDR filter is designed based on the degree to which the degree of interaural cross-correlation of noise components included in the input signal is maintained (second aspect).

More preferably, the beamforming method of the second aspect further includes an adjustment step that allows the degree to be changed (third aspect).

More preferably, in the beamforming method of the third aspect, the cost function of the MVDR filter is expressed by an equation including a parameter that controls the degree, and in the adjustment step, the value of the parameter can be changed (the fourth aspect ).

Generally, when beamforming is performed using an MVDR filter to preserve the IC of the diffuse noise component, that is, when beamforming is performed using the MVDR-IC algorithm, there is a trade-off between noise suppression performance and IC retention performance. occurs.

On the other hand, in any of the above-mentioned beamforming methods, the MVDR filter has a predetermined design, more specifically, the degree of IC retention of the noise component contained in the input signal (and therefore the degree of trade-off). The design has been made. Moreover, the degree, more specifically, the value of the parameter that controls the degree, can be changed. Therefore, according to any of the beamforming methods described above, the user can adjust the parameters as appropriate while checking the sound output to the outside through the convolution process, or the parameters can be automatically adjusted according to the environment. By making adjustments, it is possible to achieve appropriate beamforming according to the listening environment.

Preferably, in the beamforming method according to the aspect described above, in the convolution step, the input signal is convolved on the first signal path. In addition, in the filter updating process, on a second signal path branched from the first signal path, a predetermined calculation is performed for each frequency band on a frequency domain signal corresponding to the input signal. A filter coefficient is calculated based on the result of multiplying by the gain (fifth aspect).

When various types of filtering in the frequency domain on the input signal is performed on the first signal path, the delay caused by input buffering until frequency analysis is performed, and the synthesis of the frequency domain signal into the time domain after filtering. The problem is that the delay that occurs when doing so increases depending on the frequency resolution.

On the other hand, in the beamforming method of the fifth aspect, various gain calculations are performed for each frequency band in the frequency domain on the second signal path branched from the first signal path, and in this process, the MVDR Gain is also applied.
Since coefficients based on the results are calculated on the second signal path and supplied to the FIR filter on the first signal path, according to the beamforming method of the fifth aspect, there is no delay due to analysis and synthesis. , the filtering can be accomplished by an FIR filter on the first signal path. As a result, beamforming with lower delay and more natural hearing can be achieved.

As described above, according to the present invention, appropriate beamforming can be achieved.

FIG. 1 is a block diagram schematically showing a configuration example of a binaural hearing device 100 including a binaural beamformer 1 according to an embodiment. FIG. 2 is a diagram showing an example of a basic configuration of binaural beamforming. FIG. 3 is a diagram more specifically illustrating a basic configuration example of binaural beamforming. FIG. 2 is a block diagram showing in detail a configuration example of a binaural hearing device 100 with two input channels. It is a figure showing an example of the flow of processing in a filter bank of an embodiment. FIG. 7 is a diagram (1/3) showing an example of the flow of processing in a filter bank of a comparative example. FIG. 3 is a diagram (2/3) showing an example of the flow of processing in a filter bank of a comparative example. FIG. 3 is a diagram (3/3) illustrating an example of the flow of processing in a filter bank of a comparative example. FIG. 3 is a diagram showing an example of the basic configuration of binaural beamforming when the number of input channels is increased.

Hereinafter, embodiments of the present invention will be described with reference to the drawings. Note that the following embodiments are preferred examples, and the present invention is not limited to these examples.

FIG. 1 is a block diagram schematically showing a configuration example of a binaural hearing device 100 including a binaural beamformer (binaural beam forming device, binaural beam forming system) 1 according to an embodiment.

The binaural beamformer 1 can be installed in various binaural hearing devices (for example, hearing aids, etc.) that can apply gain to each frequency band. The binaural listening device 100 includes, for example, a sound input section 10 having a plurality of channels, a signal processing section 20 having a binaural beamformer 1 and a parameter adjustment section 2, a sound output section 30 having two left and right channels, and a user interface. and an operation input section 40 that accepts operations.

The sound input section 10 is a microphone, and converts the sound input into the plurality of microphones into an electrical signal (hereinafter, this signal is referred to as an "input signal"), and sends it to the signal processing section 20. In response to this, the binaural beamformer 1 performs various signal processing including beamforming using MVDR on the input signal of each microphone, and outputs the processed signal to the sound output section 30. The MVDR-IC algorithm is applied to the binaural beamformer 1. Note that details of the MVDR filter will be described in detail later. The sound output section 30 is a microphone or a speaker, and converts the signals for the left and right channels output from the binaural beamformer 1 into sound and outputs the sound to the outside. The signal processing unit 20 can be implemented, for example, by signal processing by a processor such as a DSP (digital signal processor).

In the processing by the binaural beamformer 1, as described above, a trade-off occurs between IC retention and noise suppression. Therefore, the binaural listening device 100 is provided with a configuration for externally controlling the trade-off with respect to the binaural beamformer 1. Specifically, regarding parameters for controlling the degree of IC retention of noise components (hereinafter referred to as "trade-off parameters"), for example, multiple types of preset setting values can be set according to the user's selection. The trade-off parameters can be adjusted to suit the listening environment.

The operation input unit 40 is, for example, an operation button or a touch panel. The operation input unit 40 receives changes in the set values of the trade-off parameters in response to user operations, and notifies the parameter adjustment unit 2 of the changes. Upon receiving this notification, the parameter adjustment unit 2 changes the set value of the trade-off parameter used by the binaural beamformer 1 in the process of processing to the set value selected by the user. Thereby, the binaural beamformer 1 performs signal processing using the changed trade-off parameter, more precisely, using an equation using the trade-off parameter as the cost function of the MVDR filter.

Note that since the binaural beamformer 1 performs signal processing using trade-off parameters that can be adjusted via the parameter adjustment unit 2, the parameter adjustment unit 2 is regarded as a part of the binaural beamformer 1 (i.e., the signal It is also possible to consider the entire processing section 20 as a binaural beamformer 1).

Furthermore, the manner of adjusting the trade-off parameters described above is merely an example, and the present invention is not limited thereto. For example, instead of presetting multiple types of setting values for trade-off parameters, the trade-off parameters may be automatically adjusted according to the environment, and the adjustment may be automated by applying an adaptive algorithm or the like. It is possible. In addition, the value of the trade-off parameter is automatically changed gradually, and when the user uses the earphones to check the sound that reflects the changed trade-off parameter, and the user feels that the sound is the best, By operating the operation input unit 40, it is also possible to store and learn the value of the trade-off parameter at that time.

FIG. 2 is a diagram showing an example of the basic configuration of beam forming in the binaural beam former 1. "w _L " in FIG. 2 is an MVDR filter for left channel output, and "w _R " is an MVDR filter for right channel output.

In the case of two input channels, the input signals y _L and y _R of the two input channels are input to the filter w _L and the filter w _R , respectively, as shown in FIG. As a result of analyzing the directivity of the input signal in each of the filters wL and _wR , the filter _wL outputs a signal _zL , and the filter _wR outputs a signal _zR .

Each signal in the figure is defined by the following formula. For convenience of explanation, in the expression of the definition, the signal corresponding to the left channel is represented by a variable with the subscript "L", and the signal corresponding to the right channel is represented by a variable with the subscript "R". There is. Furthermore, a signal in which left and right channel signals are superimposed is expressed using the same variables as the signals of each channel but in a different font.

In the above formula, "y" indicates the input signal, "x" indicates the desired audio signal included in the input signal (hereinafter simply referred to as "audio signal"), and "n" indicates the desired audio signal included in the input signal. This shows the noise signal that is generated. Regarding the noise signal, " ⁿⁱ " indicates a directional noise signal and " ^nv " indicates a diffuse noise signal. In addition, "s" indicates the dry source audio signal, and "a" indicates the acoustic transfer function (hereinafter referred to as "ATF"), that is, the transmission of the desired sound from the sound source to the microphone. It shows the function. "T" indicates transposition. Based on these definitions, the cost function J _MVDR of the MVDR filter can be expressed by the following formula.

As can be seen from the above formula, the cost function J _MVDR guarantees the preservation of the audio signal. Then, solutions w _L and w _R that minimize this cost function J _MVDR are obtained by the following formulas, respectively.

Note that in the above formula, both "*" and "H" indicate conjugate transposition, and "E" indicates the expected value.

As mentioned above, "a" represents the ATF of the desired voice, but it is difficult to directly estimate this in an actual environment. Therefore, the filter coefficients are calculated using the following formula on the premise that the voice section and the noise section are known in advance.

Note that in the above formula, "N" indicates the number of input channels.

Although MVDR is an optimal filter for minimizing distortion of the audio signal, the problem is that noise signals are also perceived as coming from the same direction as the audio signal. Regarding this point, the above-mentioned non-patent document 4 states that in a diffuse noisy environment, when the desired speech component and the noise component both arrive from the same direction, the SRT corresponding to 50% speech intelligibility does not improve. It has been reported that Therefore, the binaural beamformer 1 employs an MVDR-IC that holds an IC in order to spatially separate the output audio component and the residual noise component. The cost function J of MVDR- _IC can be expressed by the following formula.

In the above formula, "λ" indicates a trade-off parameter. Further, the input side and output side ICs (IC _v ⁱⁿ , IC _V ^out ) of the diffuse noise component are respectively determined by the following formulas.

It is known that preserving the IC of the diffuse noise component is a trade-off with suppressing the noise. In this regard, the binaural beamformer 1, which is configured to be able to adjust the trade-off parameter λ from the outside, allows the user to adjust the trade-off parameter λ by himself or herself or automatically according to the listening environment. Forming can be realized more appropriately.

FIG. 3 is a diagram more specifically showing the basic configuration example shown in FIG. 2. As shown in FIG. 3, when there are two input channels, four MVDR filters are provided correspondingly.

Among the four MVDR filters, “w _LL ” and “w _LR” in FIG. 3 correspond to the MVDR filter w _L for left channel output shown in FIG. 2, and “w RL ” and “w _LR ” in FIG. “w _RR ” corresponds to the MVDR filter w _R for right channel output shown in FIG. In other words, the MVDR filter w _L for left channel output is expressed as a matrix having filters w _LL and w _LR as elements, and the MVDR filter w _R for right channel output has filters w _RL and w _RR as elements. It is represented as a matrix with . For convenience of explanation, in the following explanation, filter w _LL is referred to as "first MVDR filter", filter w _RL is referred to as "second MVDR filter", filter w _LR is referred to as "third MVDR filter", and filter w _RR is referred to as "fourth MVDR filter".
MVDR filter.

The input signal y _L of the left input channel is input to the first MVDR filter w _LL and the second MVDR filter w _RL , and the input signal y _R of the right input channel is input to the third MVDR filter w _LR and the fourth MVDR filter w _RR . , each MVDR filter outputs a result based on the directivity of the input signal. Then, the signal that has passed through the first MVDR filter w _LL and the signal that has passed through the third MVDR filter w _LR are added together and output to the left channel, and the signal that has passed through the second MVDR filter w _RL and the signal that has passed through the fourth MVDR filter w _RR are added together. output to the right channel.

FIG. 4 is a block diagram showing in detail a configuration example of a binaural listening device 100 with two input channels. Note that in order to facilitate understanding of the binaural beamformer 1, illustration of the parameter adjustment section 2 and the operation input section 40 is omitted in FIG.

The binaural listening device 100 includes two microphones 10, a binaural beamformer 1, and two earphones 30, and the binaural beamformer 1 includes, for example, two input buffers 21 and two converters. 22, two hearing aid processing units 23, four MVDR filters 24, four multiplication units 25, four inverse transformation units 26, four FIR filters 27, and two addition units 28.

When sound is input to the microphone 10, the input signal is buffered in the input buffer 21 for frequency analysis, and then the transform unit 22 performs fast Fourier transform (hereinafter referred to as "fast Fourier transform") on the input signal (time domain signal) at a desired timing. (referred to as "FFT") to generate a frequency domain signal. For the signal in the frequency domain, the hearing aid processing unit 23 performs hearing aid processing by calculating the compression gain by WDRC (wide dynamic range compression) for each frequency band, and the multiplier 25 applies the hearing aid processing to the signal after the hearing aid processing. The MVDR filter 24 is applied to the signal, and the inverse transform unit 26 performs inverse fast Fourier transform (hereinafter referred to as "IFFT") on the signal after applying the MVDR filter. The IFFT provides a time domain impulse response that takes into account the hearing aid processing gain and the MVDR filter, that is, the coefficients of the FIR filter 27. The coefficients obtained by IFFT are supplied to the FIR filter 27, and the FIR filter 27 convolves the input signal using the coefficients as coefficients.

In this way, beamforming coefficients are obtained as a result of processing performed in the frequency domain, and these coefficients are supplied to the FIR filter 27 to switch the FIR filter 27, and as a result, the beamforming coefficients are switched. . If viewed locally, the "filter update unit" that updates beamforming coefficients and switches the FIR filter 27 corresponds to the inverse transform unit 26 that calculates coefficients and supplies them to the FIR filter 27, and if viewed broadly, The components involved in the processing from frequency analysis to the supply of coefficients, that is, the conversion unit 22 related to frequency analysis, the hearing aid processing unit 23, the MVDR filter 24, the multiplication unit 25, and the coefficients are calculated based on the results of these processes. This corresponds to the inverse transformer 26 that supplies the FIR filter 27 with the FIR filter 27.

FIG. 5 is a diagram showing an example of the filter bank of the embodiment, and shows the flow from when an input signal enters the input buffer 21 to being processed by the FIR filter 27 in the embodiment. As shown in FIG. 5, in the embodiment, the frequency domain line (signal path of steps SF1 to SF4) is branched and separated from the input signal line (signal path of steps SS1 to SS2), so-called side It uses a filter bank with a branch structure. In a filter bank with a side branch configuration, frequency domain signal processing performed on a frequency domain line and time domain signal processing performed on an input signal line are performed in parallel.

On the frequency domain line, FFT is performed (step SF1), hearing aid processing is performed for each frequency band (step SF2), MVDR gain is applied (step SF3), and inverse Fourier transform is performed (step SF3). SF4). As a result, coefficients of the FIR filter are obtained in which the hearing aid processing gain determined based on the frequency analysis result and the MVDR gain are taken into account. Generally, convolution (FIR filter) in the time domain is expressed by multiplication in the frequency domain. In this configuration, since hearing aid processing is added to each frequency band, beamforming is realized by multiplying the gain to which the hearing aid processing has been added by the MVDR gain. On the other hand, on the input signal line, the buffered input signal (step SS1) is convolved with a finite impulse response (FIR), and a signal is output (step SS2).

In this way, by configuring the filter bank in a side branch configuration, beamforming multiplies the gain to which frequency domain hearing aid processing has been applied by the MVDR gain, and in parallel, applies the FIR filter to the input time domain signal. This can be achieved with a simple configuration of convolving. Therefore, since no additional delay is required to implement beamforming, the delay time that may occur during signal processing can be reduced.

FIGS. 6, 7A, and 7B are diagrams showing examples of three filter banks as comparative examples. Of these, Comparative Example 1 and Comparative Example 2 are examples of analysis/reconstruction type filter banks that do not have a side branch configuration. Comparative Example 3 shows an example of a filter bank with a side branch configuration in which a part of the embodiment is intentionally modified for the purpose of comparison only.

In Comparative Example 1, the frequency domain line (signal path from steps S3' to S5') is configured in series on the input signal line (signal path from steps S1' to S6'). This differs from the filter bank of the embodiment in that the MVDR filter is not separated from the input signal and that the MVDR filter is directly applied to the input signal (step S1').

Generally, when performing array signal processing using a plurality of microphones, it is often performed at the input stage of the system, in which case a processing delay occurs at the input stage (first delay). In the filter bank of Comparative Example 1, since the MVDR filter is applied directly to the time domain signal (step S1'), the first delay occurs here, and the larger the number of buffered samples, the larger the delay. Become.

In addition, when beamforming is implemented using a filter bank in the analysis/resynthesis system, the input signal is buffered (step S2'), frequency analyzed and processed in the frequency domain, and then the signal is converted to the time domain. Therefore, there is also a delay between analysis and synthesis (second delay). In the filter bank of Comparative Example 1, the signal after applying the MVDR filter is subjected to frequency analysis on the input signal line and converted to a time domain signal (steps S3' to S5'), and then the signal is converted into a time domain signal. Since composition is performed (step S6'), a second delay occurs here. Generally, the higher the frequency resolution, the greater the delay.

In addition, in Comparative Example 2, the MVDR filter is not applied directly to the input signal, but is applied to the filter for hearing aid processing performed in the frequency domain (S13' to S14'), so the first delay does not occur. However, since it does not have a side branch configuration, it is necessary to transform the frequency domain signal into the time domain and synthesize it, so the second delay is unavoidable.

In Comparative Example 3, since the side branch configuration is adopted, the second delay is suppressed within the range that can occur due to frequency analysis itself, but the MVDR filter is not directly applied to the input signal. (SS1'), the first delay is unavoidable.

As described above, in the filter bank of the comparative example, occurrence of a large delay due to the above two factors is unavoidable.

On the other hand, in the filter bank of the embodiment, the MVDR filter is applied on the line in the frequency domain, so the first delay does not occur. Further, among the second delays, only a slight delay caused by the frequency analysis itself does not occur. This is because in the filter bank of the embodiment, the frequency domain line is separated from the signal input line, so the coefficients obtained based on the frequency analysis results can be supplied to the FIR filter section and reflected. This is because the process of converting a signal from the frequency domain to the time domain as in the comparative example is not necessary.

Therefore, according to the filter bank of the embodiment, the delay time can be significantly shortened compared to the filter bank of the comparative example, so that natural hearing can be achieved. Furthermore, the amount of calculation required for processing is reduced, and power consumption can be reduced.

[See Figure 4: MVDR filter 24]
Further, as shown in FIG. 4, the binaural beamformer 1 has four MVDR filters 24 corresponding to two input channels (left microphone 10-1, right microphone 10-2). Specifically, the left channel input signal y _L is input to the first MVDR filter 24-1 (w _LL ), and the input signal y L of the left channel is input to the second MVDR filter 24-2 (w _RL ) and the third MVDR filter 24-3 (w _LR ), respectively. The left channel and right channel input signals y _L and y _R are input to the fourth MVDR filter 24-4 (w RR ), and the right channel input signal y _R is input to the fourth MVDR filter 24-4 (w _RR ).

Then, the input signal _yL of the left channel is convoluted with the hearing aid processing in the first hearing aid processing unit 23-1 for the left input channel and the first FIR filter 27-1 in which the coefficients of the first MVDR filter 24-1 are added. , the input signal _yL of the left channel is convoluted with the hearing aid processing in the first hearing aid processing unit 23-1 for the left channel and the second FIR filter 27-2, which takes into account the coefficients of the second MVDR filter 24-2. The input signal yR of the right channel is convoluted with the third FIR filter 27-3 in which the hearing aid processing in the second hearing aid processing unit 23-2 for the channel and the coefficients of the third MVDR filter 24-3 are added, and the input signal _yR of the right channel is convoluted. The input signal _yR of the right channel is convolved with the fourth FIR filter 27-4, which takes into account the hearing aid processing performed by the second hearing aid processing unit 23-2 and the coefficients of the fourth MVDR filter 24-4.

Then, the signals convoluted by the first FIR filter 27-1 and the third FIR filter 27-3 are added by the first adder 28-1 and output to the left earphone 30-1. This will output sound on the left channel. Furthermore, the signals convoluted by the second FIR filter 27-2 and the fourth FIR filter 27-4 are added by the second adder 28-2 and output to the right earphone 30-2. This will output sound on the right channel.

By adopting the above configuration, the binaural beamformer 1 can highlight the desired audio signal while appropriately suppressing the noise signal included in the input signal, taking into account the directivity of the input signal. As a result, a state in which the desired audio signal appears to be emphasized can be obtained, making it possible to make the desired audio sound more natural and easier to hear.

[Advantages of the present invention]
As described above, according to the embodiment described above, the following effects can be obtained.

In the binaural beamformer 1, the trade-off parameters are configured to be adjustable from the outside, so the trade-off parameters can be selected externally according to the environment and the trade-off between IC retention performance and noise suppression performance can be adjusted externally. Can be adjusted. For example, the user of the binaural beamformer 1 can adjust the trade-off parameter as appropriate while checking the sound output from the left and right earphones 30-1 and 30-2. As a result, more appropriate beamforming can be achieved depending on the listening environment.

A filter bank with a side branch configuration is used, and processing in the frequency domain is performed in parallel with processing in the time domain, and beamforming is performed by multiplying the gain in the frequency domain by the gain of the MVDR, so the analysis・Compared to the case of using a filter bank that performs reconstruction, the amount of calculations that occur during the processing process is smaller, and the delay can be significantly reduced. As a result, beamforming can be achieved with low delay, and more natural hearing can be achieved.

The present invention is not limited to the embodiments described above, and can be implemented with various modifications.

In the embodiment described above, the input is configured with two channels, but the number of input channels can be increased to an arbitrary number N.

FIG. 8 is a diagram showing an example of the basic configuration of binaural beamforming when the number of input channels is N. In this case, a necessary number of multipliers, inverse transformers, FIR filters, etc. may be provided according to the number of input channels so as to have the configuration shown below. 2M=N (M, N are natural numbers) microphone input signals y _L1 , y _L2 , . . . , y to each of the MVDR filter w _L for left channel output and the MVDR filter w _R for right channel output _LM is input. As a result of analyzing the directivity of the input signal in each of the filters w _L and w _R , a signal z _L is output from the filter w _L. MVDR filters (corresponding to the plurality of MVDR filters shown in FIGS. 3 and 4) are provided as elements of each matrix of the MVDR filters w _L and w _R so that the signal z _R is output from the filter w _R.

In addition, the configurations, numerical values, etc. mentioned in the process of explanation regarding the binaural beamformer 1 and the binaural listening device 100 are merely examples, and it goes without saying that they can be modified as appropriate when implementing the present invention.

This application is based on Japanese Patent Application No. 2022-076676 filed on May 6, 2022, the contents of which are incorporated herein by reference.

1 Binaural beamformer 2 Parameter adjustment section (adjustment process, adjustment section)
10 Microphone 20 Signal processing unit 21 Input buffer 22 Conversion unit (filter update process, filter update unit)
23 Hearing aid processing unit (filter update process, filter update unit)
24 MVDR filter (filter update process, filter update section)
25 Multiplication unit (filter update process, filter update unit)
26 Inverse transformation unit (filter update process, filter update unit)
27 FIR filter (convolution process, convolution section)
28 Addition unit 30 Earphone 100 Binaural listening device

Claims (10)

1. A beamforming method that performs beamforming using MVDR on input signals corresponding to sounds input to multiple microphones, the method comprising:
   a filter updating step of calculating coefficients based on the result of passing the input signal through an MVDR filter having a predetermined design, and switching an FIR filter using the coefficients;
   and a convolution step of convolving the input signal with the FIR filter.
2. The beamforming method according to claim 1,
   The MVDR filter is
   It is characterized in that it is designed based on the degree to which interaural cross-correlation of noise components included in the input signal is maintained.
3. The beamforming method according to claim 2,
   The method is characterized in that it further includes an adjustment step that allows the degree to be changed.
4. The beamforming method according to claim 3,
   The MVDR filter is
   The cost function is expressed by an expression including a parameter that controls the degree,
   In the adjustment step,
   The method is characterized in that the value of the parameter can be changed.
5. The beamforming method according to any one of claims 1 to 4,
   In the convolution process,
   convolving the input signal on a first signal path;
   In the filter updating step,
   On a second signal path branched from the first signal path, a predetermined calculation is performed for each frequency band on a frequency domain signal corresponding to the input signal, and the frequency domain gain is multiplied by the MVDR gain. The method is characterized in that the coefficients are calculated based on the results.
6. A beamforming system that uses MVDR to perform beamforming on input signals corresponding to sounds input to multiple microphones, the system comprising:
   a filter updating unit that calculates coefficients based on the result of passing the input signal through an MVDR filter having a predetermined design, and switches an FIR filter using the coefficients;
   and a convolution unit that convolves the input signal with the FIR filter.
7. The beamforming system according to claim 6,
   The MVDR filter is
   It is characterized in that it is designed based on the degree to which interaural cross-correlation of noise components included in the input signal is maintained.
8. The beamforming system according to claim 7,
   The invention is characterized in that it further includes an adjustment section that allows the degree to be changed.
9. The beamforming system according to claim 8,
   The MVDR filter is
   The cost function is expressed by an expression including a parameter that controls the degree,
   The adjustment section is
   The method is characterized in that the value of the parameter can be changed.
10. The beamforming system according to any one of claims 6 to 9,
    The convolution part is
    convolving the input signal on a first signal path;
    The filter updating unit includes:
    On a second signal path branched from the first signal path, a predetermined calculation is performed for each frequency band on a frequency domain signal corresponding to the input signal, and the frequency domain gain is multiplied by the MVDR gain. The method is characterized in that the coefficients are calculated based on the results.

Images