Processing method for removing late reflection reverberation of microphone array

Abstract

The invention discloses a processing method for removing late reflection reverberation of a microphone array, which relates to the technical field of voice signal processing, and comprises the following steps: a fixed beam forming module performs spatial filtering processing on an original signal by using a specified beam to obtain a first voice signal; the noise reduction and reverberation removal module generates an attenuation gain mask according to the first voice signal, and performs noise reduction processing on the first voice signal by using the mask to obtain a second voice signal; the voice direction estimation module carries out weighted calculation on the original signal according to the mask and carries out voice direction estimation to obtain a target voice direction; the fixed beam forming module updates a specified beam according to the target voice direction; and the early reverberation simulation module performs reverberation addition processing on the second voice signal and outputs a final voice signal. Optimization of fixed beam processing is realized through voice estimation; and after the reverberation is eliminated, reverberation adding processing is performed on the signal in a space reverberation simulation mode, and early reflection is simulated, so that the sound is clearer and fuller.

Description

Technical Field

[0001] The present invention relates to the technical field of speech signal processing, and in particular to a method for removing late reflection reverberation of a microphone array. Background Art

[0002] In a conference call, it is very important to get clear voice quality, especially when the distance is long, and there are challenges such as low signal-to-noise ratio and high reverberation, which will seriously affect the clarity of the call. At present, the reverberation suppression technology for conference scenes generally has the following shortcomings:

[0003] 1. Traditional Wiener filtering or microphone array methods often lead to distortion of the main speech signal or poor reverberation suppression effect;

[0004] 2. Although many deep learning-based dereverberation technologies can effectively suppress reverberation, excessive suppression may eliminate early reverberation that is beneficial to human hearing, making the target speech sound unnatural and lacking in liveliness;

[0005] During a call, voice reverberation is mainly caused by the reflection of sound waves. This reflection can be divided into early reflection and late reflection. Early reflection can enhance the fullness and clarity of the sound, while late reflection will make the voice muddy and unclear. Traditional reverberation processing technology often has difficulty distinguishing between these two types of reflections and usually treats them equally. The result is either that the reverberation is not effectively suppressed and the call quality is still poor; or that the reverberation is excessively eliminated, causing the voice to sound dry, unnatural, or even distorted. Summary of the invention

[0006] In view of the problems existing in the prior art, the present invention provides a method for removing late reflection reverberation of a microphone array to solve the problem of voice quality degradation caused by conference scenarios, especially the far field of long-distance calls.

[0007] The present invention uses a microphone array algorithm combined with deep learning technology to perform fine processing on reverberation. By simulating reverberation and restoring early reverberation, the interference of reverberation on speech is suppressed, and the clarity and fullness of speech are enhanced, thereby providing a better call experience. The technical solution is implemented as follows:

[0008] A method for removing late reflection reverberation of a microphone array, comprising a fixed beam forming module, a noise reduction and reverberation removal module, a speech direction estimation module and an early reverberation simulation module; and also comprising a microphone array composed of a plurality of microphones; the microphone array is used to collect original signals;

[0009] The following steps are also included:

[0010] S1, the fixed beamforming module receives the original signal, and uses the specified beam to perform spatial filtering on the original signal to obtain a first speech signal, that is, using the beam to suppress a part of the reverberation, that is, performing beam enhancement; sending the first speech signal to the noise reduction and dereverberation module; sending the original signal to the speech direction estimation module;

[0011] The beam coefficient is the filter coefficient configured for the array microphone. This coefficient affects the sound pickup direction. It is a beam designed according to the sound pickup target direction. The beam is used to enhance the sound in the target direction angle.

[0012] S2, the noise reduction and dereverberation module generates an attenuation gain mask according to the first speech signal, and uses the attenuation gain mask to perform noise reduction processing on the first speech signal to obtain a second speech signal;

[0013] The attenuation gain mask is sent to the speech direction estimation module, and step S3 is executed;

[0014] The second speech signal is sent to the early reverberation simulation module, and step S5 is executed;

[0015] S3, the speech direction estimation module performs weighted calculation on the original signal according to the attenuation gain mask, and estimates the speech direction to obtain the target speech direction; sends the target speech direction to the fixed beamforming module, and executes step S4;

[0016] S4, the fixed beam forming module selects and updates the designated beam according to the target speech direction; that is, selects a beam in a new direction and uses the designated beam for spatial filtering;

[0017] S5. The early reverberation simulation module performs reverberation processing on the second speech signal and outputs a final speech signal.

[0018] The original signal is also the array signal. The fixed beamforming module is fixed beamforming. Through the noise reduction and dereverberation processing of the fixed beamforming module, the noise reduction and dereverberation module, the obtained speech signal, especially the speech signal at a long distance, is distorted, affecting the hearing of the speech; the early reverberation simulation module mirrors the speech signal for early reflection recovery processing to restore the naturalness and saturation of the speech; the direction of the speech signal is located through speech direction estimation, and the target speech direction obtained can further optimize the beam processing and dynamically adjust the beam direction, that is, update the specified beam, so as to process the speech signal more accurately.

[0019] As a further optimization of the above solution, the number of the microphones is 8 or 6; the fixed beam forming module fixes the designated beam through a preset filter coefficient.

[0020] The number of microphones may also be other numbers, preferably 8 or 6.

[0021] As a further optimization of the above solution, the weighted calculation is to perform time-frequency weighting on the second speech signal through the attenuation gain mask, that is:

[0022]

[0023] Wherein, X is the second signal, and X(l,k) is the frequency value of the kth frequency point in the lth frame;

[0024] mask(l,k) is the attenuation gain value corresponding to the kth frequency point in the lth frame;

[0025] Multiply X(l,k) and mask(l,k) to get the time-frequency point

[0026] In speech direction estimation, in order to obtain the denoised data of each microphone channel, the following formulas are used to calculate:

[0027]

[0028] Among them, X m (l, k) is the frequency value of the microphone at the kth frequency point in the mth channel and the lth frame; one microphone corresponds to one channel.

[0029] The speech direction estimation module is based on the time-frequency point The speech direction is estimated using the doa algorithm.

[0030] The speech signal has corresponding spectral data in each frame, and the frequency points can be used to identify specific phonemes, tones or other features in the speech. For example, the fundamental frequency (F0) is an important frequency point in the speech signal, which is directly related to the pitch of the speaker.

[0031] The direction of arrival (DOA) estimation algorithm is a key technology for determining the direction of a signal source and is widely used in radar, sonar, wireless communication and other fields. The doa estimation algorithm used in this solution can be music (Multiple Signal Classification), srp-phat (Steered-Response Power PHAT), dsb (Delay and Sum Beamforming), etc., but is not limited to these methods.

[0032] X m (l,k) and The mask is used to perform a time-frequency weighting on the multi-channel spectrum data to remove the interference of non-target speech signals. The advantage is that a very good speech localization effect can still be obtained when there is a lot of noise in the calculation environment.

[0033] As a further optimization of the above solution, the noise reduction and dereverberation module is a neural network structure based on CRNN;

[0034] The training process of the neural network structure includes inputting clean data and first noise data;

[0035] adding reverberation to the clean data and the first noise data to obtain second noise data;

[0036] Inputting the amplitude spectrum or log spectrum of the second noise data as a feature into the neural network structure, and outputting an attenuation gain mask of the amplitude spectrum;

[0037] Using the attenuation gain mask to reduce noise on the second noise data to obtain an enhanced speech signal;

[0038] The enhanced speech signal and the clean data are combined with an L1 loss function to calculate and optimize the neural network structure.

[0039] CRNN, namely Convolutional Recurrent Neural Network, is also a convolutional recurrent neural network, which integrates convolutional neural network (CNN) and recurrent neural network (RNN); it includes an encoding layer, an intermediate layer and a decoding layer; the encoding layer adopts a CNN convolution operator; the intermediate layer adopts a GRU recursive operator; the decoding layer adopts a CNN-transpose deconvolution operator.

[0040] In the inference phase, the network parameters of the training process are directly copied, and the network is transformed into a streaming inference mode. The data output by beamforming is input and denoised.

[0041] As a further optimization of the above solution, the L1 loss function is:

[0042] L1 loss=mean(abs(abs(Y(l,k))×Mask(l,k)-abs(X(l,k))));

[0043] Among them, Mask(l,k) represents the attenuation gain mask; X(l,k) represents the clean data; Y(l,k) represents the second noise data, abs(Y(l,k))×Mask(l,k) represents the enhanced speech signal, and the same weighted calculation method as above is used here; the mean() function is used to calculate the average of all differences to obtain the L1 loss; abs() is the absolute value function; (l,k) represents the kth frequency point of the lth frame.

[0044] As a further optimization of the above solution, the early reverberation simulation module includes a plurality of comb filters connected in parallel and a plurality of all-pass filters connected in series thereafter;

[0045] The comb filters perform filtering processing on the second speech signals respectively; the input results of multiple comb filters are added to obtain a third speech signal;

[0046] The third voice signal is sent to the all-pass filter for filtering in sequence to obtain a fourth voice signal;

[0047] The fourth voice signal, the second voice signal and the preset gain signal are superimposed to obtain the final voice signal.

[0048] As a further optimization of the above scheme, the filtering process of the comb filter is expressed as:

[0049] y(t)=x(t-τ)+g×y(t-τ);

[0050] The filtering process of the all-pass filter is expressed as:

[0051] y(t)=-g×x(t)+x(t-τ)+g×y(t-τ);

[0052] Among them, x(·) is the input signal, y(·) is the output signal, t is the time, τ is the preset delay; g is the preset gain coefficient.

[0053] The calculation of y(t) depends on the value of y(t-τ) that has been calculated. g is the reflection attenuation gain, that is, the attenuation of the reflected speech signal after a certain time.

[0054] As a further optimization of the above scheme, the processing of the original signal, the first speech signal, the second speech signal and the final speech signal are all in frames; it also includes a preset estimation cycle, and the interval of one estimation cycle is several frames; in step S2, after each estimation cycle, the attenuation gain mask is sent to the speech direction estimation module, and step S3 is executed.

[0055] The estimation period is 1 frame, 2 frames or other number of frames. In speech signal processing, since speech has short-term stability, that is, the characteristics of speech signals do not change much in a short time, speech signals are divided into shorter frames for processing.

[0056] As a further optimization of the above solution, the fixed beam forming module is preset with a beams, the angle difference between two adjacent beams is b; a×b=360; the central angle of one beam is 0°, and the degrees of the remaining central angles are all integer multiples of b.

[0057] As a further optimization of the above solution, the target speech direction is an angle; in step S4, a beam whose center angle is closest to the target speech direction is selected as the designated beam.

[0058] For example, if the current target speaker angle (target speech direction) is recognized to be 40 degrees, a beam with a center angle of 30 degrees will be selected as the designated beam.

[0059] Compared with the prior art, the present invention achieves the following beneficial effects:

[0060] (1) The noise reduction and dereverberation processing of the fixed beamforming module, noise reduction and dereverberation module can eliminate most of the interference of reverberation. For defects such as speech and distortion caused by reverberation elimination, the reverberation signal is reverberated by simulating the early reflection in the reverberation to make the sound clearer and fuller.

[0061] (2) The direction of the speech signal is located by estimating the speech direction. The target speech direction obtained can be used to further optimize the beam processing and dynamically adjust the beam direction, that is, update the specified beam, so as to process the speech signal more accurately.

[0062] (3) By using masked a posteriori estimation, the multi-channel spectral data is weighted by time-frequency points through masking to remove interference from non-target speech signals. This allows very good speech localization results to be calculated even in the presence of very large external noise. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 It is a data flow diagram of a method for removing late reflection reverberation of a microphone array provided by an embodiment of the present invention;

[0064] Figure 2 is a schematic diagram of beam pointing provided by an embodiment of the present invention;

[0065] Figure 3 is a schematic diagram of a neural network structure provided by an embodiment of the present invention;

[0066] Figure 4It is a schematic diagram of the data processing flow of the early reverberation simulation module provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0067] In order to make the purpose, technical solution and advantages of the present invention more clear, the technical solution in the embodiment of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiment of the present invention. Obviously, the described embodiment is only a part of the embodiment of the present invention, not all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0068] like Figures 1 to 4 As shown, this embodiment provides a method for removing late reflection reverberation of a microphone array, including a fixed beamforming module, a noise reduction and reverberation removal module, a speech direction estimation module and an early reverberation simulation module.

[0069] It also includes a microphone array composed of multiple microphones; in this embodiment, the number of microphones is 8. The microphone array is used to collect the original signal;

[0070] The following steps are also included:

[0071] S1, the fixed beamforming module receives the original signal, and uses the specified beam to perform spatial filtering on the original signal to obtain the first speech signal, that is, using the beam to suppress part of the reverberation; the first speech signal is sent to the noise reduction and dereverberation module; the original signal is sent to the speech direction estimation module;

[0072] The beam coefficient is the filter coefficient configured for the array microphone. This coefficient affects the sound pickup direction. It is a beam designed based on the target sound pickup direction. The beam is used to enhance the sound in the target direction angle.

[0073] The fixed beam forming module fixes the specified beam through the preset filter coefficients.

[0074] In this embodiment, the fixed beam forming module is preset with 12 beams, and the angle difference between two adjacent beams is 30°; the central angle of one beam is 0°, and the degrees of the remaining central angles are integer multiples of 30°. The central angles of each beam are: 0°, 30°, 60°, 90°, 120°, 150°, 180°, 210°, 240°, 270°, 300°, 330°.

[0075] Figure 2 In the provided directional diagram, the direction is at the 180° position and the width is 60°.

[0076] S2, the noise reduction and dereverberation module generates an attenuation gain mask according to the first speech signal, and uses the attenuation gain mask to perform noise reduction processing on the first speech signal to obtain a second speech signal.

[0077] In this embodiment, the noise reduction and dereverberation module is a CRNN-Network based neural network structure;

[0078] The training process of the neural network structure includes inputting clean data clean and first noise data noise;

[0079] The clean data clean and the first noise data noise are reverberated by RIR (Room Impulse Response, simulating the generation of room acoustic impulse response) to obtain the second noise data;

[0080] Inputting the amplitude spectrum or log spectrum of the second noise data into the neural network structure as a feature, and outputting the attenuation gain mask of the amplitude spectrum;

[0081] Denoising the second noise data using the attenuation gain mask to obtain an enhanced speech signal;

[0082] The enhanced speech signal and clean data are combined with the L1 loss function to calculate and optimize the neural network structure.

[0083] In this embodiment, the L1 loss function is:

[0084] L1 loss=mean(abs(abs(Y(l,k))×Mask(l,k)-abs(X(l,k))));

[0085] Among them, Mask(l,k) represents the attenuation gain mask; X(l,k) represents the clean data; Y(l,k) represents the second noise data, abs(Y(l,k))×Mask(l,k) represents the enhanced speech signal, and the same weighted calculation method as mentioned above is used here; the mean() function is used to calculate the average of all differences to obtain the L1 loss; abs() is the absolute value function; (l,k) represents the kth frequency point of the lth frame.

[0086] CRNN, namely Convolutional Recurrent Neural Network, is a convolutional recurrent neural network that combines convolutional neural network (CNN) and recurrent neural network (RNN). It includes encoding layer, intermediate layer and decoding layer. The encoding layer uses the CNN convolution operator, the intermediate layer uses the GRU recursive operator, and the decoding layer uses the CNN-transpose deconvolution operator.

[0087] In the inference phase, the network parameters of the training process are directly copied, and the network is transformed into a streaming inference mode. The data output by beamforming is input and denoised.

[0088] The attenuation gain mask is sent to the speech direction estimation module, and step S3 is executed;

[0089] The second speech signal is sent to the early reverberation simulation module, and step S5 is executed.

[0090] S3, the speech direction estimation module performs weighted calculation on the original signal according to the attenuation gain mask, and estimates the speech direction to obtain the target speech direction;

[0091] In this embodiment, the weighted calculation is to perform time-frequency weighting on the second speech signal by attenuating the gain mask, that is:

[0092]

[0093] Wherein, X is the second signal, and X(l,k) is the frequency value of the kth frequency point in the lth frame;

[0094] mask(l,k) is the attenuation gain value corresponding to the kth frequency point in the lth frame;

[0095] Multiply X(l,k) and mask(l,k) to get the time-frequency point

[0096] In speech direction estimation, in order to obtain the denoised data of each microphone channel, the following formulas are used to calculate:

[0097]

[0098] Among them, X m (l,k) is the frequency value of the microphone at the kth frequency point in the mth channel and the lth frame; one microphone corresponds to one channel.

[0099] The speech direction estimation module is based on the time-frequency point And doa algorithm for speech direction estimation.

[0100] The speech signal has corresponding spectral data in each frame, and the frequency points can be used to identify specific phonemes, tones or other features in the speech. For example, the fundamental frequency (F0) is an important frequency point in the speech signal, which is directly related to the pitch of the speaker.

[0101] The direction of arrival (DOA) estimation algorithm is a key technology for determining the direction of a signal source and is widely used in radar, sonar, wireless communication and other fields. The doa estimation algorithm used in this solution can be music (Multiple Signal Classification), srp-phat (Steered-Response Power PHAT), dsb (Delay and Sum Beamforming), etc., but is not limited to these methods.

[0102] X m (l,k) and The mask is used to perform a time-frequency weighting on the multi-channel spectrum data to remove the interference of non-target speech signals. The advantage is that a very good speech localization effect can still be obtained when there is a lot of noise in the calculation environment.

[0103] The target speech direction is sent to the fixed beamforming module, and step S4 is executed; in this embodiment, the target speech direction is an angle.

[0104] S4, the fixed beam forming module selects and updates the designated beam according to the target speech direction, that is, selects a beam in a new direction, and uses the designated beam for spatial filtering. In this embodiment, the beam whose center angle is closest to the target speech direction is selected as the designated beam. For example, if it is recognized that the current target speaker angle (target speech direction) is 40 degrees, a beam with a center angle of 30 degrees will be selected as the designated beam.

[0105] S5. The early reverberation simulation module performs reverberation processing on the second speech signal and outputs a final speech signal.

[0106] In this embodiment, the early reverberation simulation module includes four comb filters connected in parallel and two all-pass filters connected in series thereafter;

[0107] The comb filters perform filtering processing on the second speech signal respectively; the input results of multiple comb filters are added together to obtain a third speech signal;

[0108] The third voice signal is sent to the all-pass filter for filtering in sequence to obtain a fourth voice signal;

[0109] The fourth voice signal, the second voice signal and the preset gain signal are superimposed to obtain a final voice signal.

[0110] Figure 4 In the figure, x(n) is the second speech signal, g_w is the preset gain signal; x'(n) is the signal obtained by superimposing the fourth speech signal and g_w, and y(n) is the final speech signal.

[0111] In this embodiment, the filtering process of a single comb filter is expressed as:

[0112] y(t)=x(t-τ)+g×y(t-τ);

[0113] The filtering process of a single all-pass filter is expressed as:

[0114] y(t)=-g×x(t)+x(t-τ)+g×y(t-τ);

[0115] Among them, x(·) is the input signal, y(·) is the output signal, t is the time, τ is the preset delay; g is the preset gain coefficient.

[0116] The calculation of y(t) depends on the value of y(t-τ) that has been calculated. g is the reflection attenuation gain, that is, the attenuation of the reflected speech signal after a certain time.

[0117] The original signal is also the array signal. The fixed beamforming module is fixed beamforming. Through the noise reduction and dereverberation processing of the fixed beamforming module, the noise reduction and dereverberation module, the obtained speech signal, especially the speech signal at a long distance, is distorted, affecting the hearing of the speech; the early reverberation simulation module mirrors the speech signal for early reflection recovery processing to restore the naturalness and saturation of the speech; the direction of the speech signal is located through speech direction estimation, and the target speech direction obtained can further optimize the beam processing and dynamically adjust the beam direction, that is, update the specified beam, so as to process the speech signal more accurately.

[0118] In this embodiment, the processing of the original signal, the first speech signal, the second speech signal and the final speech signal are all in frames; a preset estimation cycle is also included, and the interval of one estimation cycle is 10 frames; in step S2, after each estimation cycle, the attenuation gain mask is sent to the speech direction estimation module, and step S3 is executed. That is, the speech direction estimation is performed once every 10 frames.

[0119] In speech signal processing, since speech has short-term stationarity, that is, the characteristics of the speech signal do not change much in a very short period of time, the speech signal is divided into shorter frames for processing.

[0120] According to the disclosure and teaching of the above description, those skilled in the art to which the present invention belongs may also make changes and modifications to the above embodiments. Therefore, the present invention is not limited to the specific embodiments disclosed and described above, and some modifications and changes to the present invention should also fall within the scope of protection of the claims of the present invention. In addition, although some specific terms are used in this specification, these terms are only for the convenience of description and do not constitute any limitation to the present invention.

Claims

1. A method for removing late reflection reverberation of a microphone array, characterized in that: It includes a fixed beam forming module, a noise reduction and dereverberation module, a speech direction estimation module and an early reverberation simulation module; it also includes a microphone array composed of multiple microphones; the microphone array is used to collect original signals; The following steps are also included: S1, the fixed beamforming module receives the original signal, and uses the specified beam to perform spatial filtering on the original signal to obtain a first speech signal; sends the first speech signal to the noise reduction and dereverberation module; sends the original signal to the speech direction estimation module; S2, the noise reduction and dereverberation module generates an attenuation gain mask according to the first speech signal, and uses the attenuation gain mask to perform noise reduction processing on the first speech signal to obtain a second speech signal; The attenuation gain mask is sent to the speech direction estimation module, and step S3 is executed; The second speech signal is sent to the early reverberation simulation module, and step S5 is executed; S3, the speech direction estimation module performs weighted calculation on the original signal according to the attenuation gain mask, and estimates the speech direction to obtain the target speech direction; sends the target speech direction to the fixed beamforming module, and executes step S4; S4, the fixed beam forming module selects and updates the designated beam according to the target speech direction; S5. The early reverberation simulation module performs reverberation processing on the second speech signal and outputs a final speech signal.

2. The method for removing late reflection reverberation of a microphone array according to claim 1, characterized in that: The number of the microphones is 8 or 6; the fixed beam forming module fixes the designated beam through a preset filter coefficient.

3. The method for removing late reflection reverberation of a microphone array according to claim 1, characterized in that: The weighted calculation is to perform time-frequency weighting on the second speech signal through the attenuation gain mask, that is: Wherein, X is the second signal, and X(l,k) is the frequency value of the kth frequency point in the lth frame; mask(l,k) is the attenuation gain value corresponding to the kth frequency point in the lth frame; Multiply X(l,k) and mask(l,k) to get the time-frequency point The speech direction estimation module is based on the time-frequency point The speech direction is estimated using the doa algorithm.

4. The method for removing late reflection reverberation of a microphone array according to claim 3, characterized in that: The noise reduction and dereverberation module is a neural network structure based on CRNN; The training process of the neural network structure includes inputting clean data and first noise data; adding reverberation to the clean data and the first noise data to obtain second noise data; Inputting the amplitude spectrum or log spectrum of the second noise data as a feature into the neural network structure, and outputting an attenuation gain mask of the amplitude spectrum; Using the attenuation gain mask to reduce noise on the second noise data to obtain an enhanced speech signal; The enhanced speech signal and the clean data are combined with an L1 loss function to calculate and optimize the neural network structure.

5. The method for removing late reflection reverberation of a microphone array according to claim 4, characterized in that: The L1 loss function is: L1 loss=mean(abs(abs(Y(l,k))×Mask(l,k)-abs(X(l,k)))); Among them, Mask(l,k) represents the attenuation gain mask; X(l,k) represents the clean data; Y(l,k) represents the second noise data, abs(Y(l,k))×Mask(l,k) represents the enhanced speech signal; the mean() function is used to calculate the average of all differences; abs() is the absolute value function; (l,k) represents the kth frequency point of the lth frame.

6. The method for removing late reflection reverberation of a microphone array according to claim 1, characterized in that: The early reverberation simulation module includes a plurality of comb filters connected in parallel and a plurality of all-pass filters connected in series thereafter; The comb filters perform filtering processing on the second speech signals respectively; the input results of multiple comb filters are added to obtain a third speech signal; The third voice signal is sent to the all-pass filter for filtering in sequence to obtain a fourth voice signal; The fourth voice signal, the second voice signal and the preset gain signal are superimposed to obtain the final voice signal.

7. The method for removing late reflection reverberation of a microphone array according to claim 6, characterized in that: The filtering process of the comb filter is expressed as: y(t)=x(t-τ)+g×y(t-τ); The filtering process of the all-pass filter is expressed as: y(t)=-g×x(t)+x(t-τ)+g×y(t-τ); Among them, x(·) is the input signal, y(·) is the output signal, t is the time, τ is the preset delay; g is the preset gain coefficient.

8. The method for removing late reflection reverberation of a microphone array according to claim 1, characterized in that: The processing of the original signal, the first speech signal, the second speech signal and the final speech signal are all in frames; and a preset estimation cycle is also included, and the interval of one estimation cycle is a number of frames; In step S2, after each estimation cycle, the attenuation gain mask is sent to the speech direction estimation module, and step S3 is executed.

9. The method for removing late reflection reverberation of a microphone array according to claim 1, characterized in that: The fixed beam forming module is preset with a beams, the angle difference between two adjacent beams is b; a×b=360; the central angle of one beam is 0°, and the degrees of the remaining central angles are all integer multiples of b.

10. The method for removing late reflection reverberation of a microphone array according to claim 9, characterized in that: The target speech direction is an angle; in step S4, a beam whose central angle is closest to the target speech direction is selected as the designated beam.

Images