A noise elimination method, system, device and medium based on bipolar Kalman

Abstract

The present application relates to the field of noise cancellation technology, and in particular to a noise cancellation method, system, device and medium based on bipolar Kalman, comprising obtaining a total Kalman gain, updating a first Kalman filter model according to the total Kalman gain; performing a first-level Kalman adaptive processing on an array microphone signal according to the first Kalman filter model to obtain a first result signal that eliminates a common echo; obtaining a single-microphone Kalman gain, updating a second Kalman filter model according to the single-microphone Kalman gain; and performing a second-level Kalman adaptive processing on a corresponding single microphone signal according to the second Kalman filter model to obtain a second result signal that eliminates residual echo and reverberation, thereby reducing the cost of noise cancellation and improving audio quality.

Description

Technical Field

[0001] The present application relates to the technical field of noise elimination, and in particular to a noise elimination method, system, device and medium based on bipolar Kalman. Background Art

[0002] In real-time communications, echo cancellation and beamforming are crucial technologies for improving audio quality. Echo cancellation algorithms use adaptive filters to model the dynamically changing acoustic echo path from the speaker to the microphone array and then filter out the echo; beamforming algorithms use spatial filters to enhance the extraction of sound sources in the observation direction, thereby suppressing non-target sound sources in other directions. However, integrating echo cancellation and multi-microphone beamforming algorithms leads to interdependence issues due to their time-varying nature. For example, if beamforming is performed before echo cancellation, if the target sound source moves, the beamforming direction changes, potentially disrupting the echo cancellation effect, and vice versa.

[0003] At the same time, reverberation is also a factor that affects audio quality. Reverberation is the noise that is picked up by the microphone after the sound waves are reflected by walls or surrounding obstacles. It not only affects the call quality, but also affects the effects of echo cancellation and beamforming. Therefore, reverberation is also a factor that affects the sound quality received by the array microphone.

[0004] Current technology is not yet able to effectively combine echo cancellation, beamforming, and reverberation cancellation, which will affect the communication audio quality. If the original signal of the array microphone is echo cancelled and then beamformed and reverberated, it may introduce excessive computational complexity, resulting in increased computing costs. The above problems need to be solved. Summary of the Invention

[0005] In order to reduce the computational cost of noise cancellation and improve audio quality, this application provides a noise cancellation method, system, device, and medium based on bipolar Kalman filtering, which adopts the following technical solutions:

[0006] In a first aspect, the present application provides a noise elimination method based on bipolar Kalman, comprising:

[0007] Get array microphone signal;

[0008] Obtaining a total Kalman gain, and updating the total Kalman gain to obtain a first Kalman filter model;

[0009] Performing a first-level Kalman adaptive processing on the array microphone signal according to the first Kalman filter model to obtain a first result signal that eliminates the common echo;

[0010] Obtaining a single-microphone Kalman gain, and updating a second Kalman filter model according to the single-microphone Kalman gain;

[0011] A single microphone signal is obtained according to the first result signal, and a second-level Kalman adaptive processing is performed on the corresponding single microphone signal according to the second Kalman filter model to obtain a second result signal in which residual echo and reverberation are eliminated.

[0012] Preferably, the specific steps of obtaining the total Kalman gain are:

[0013] The covariance of the total Kalman measurement error at the current moment and the covariance of the total Kalman measurement error at the previous moment are obtained, and the total Kalman gain is calculated according to the covariance of the total Kalman error at the current moment and the covariance of the total Kalman error at the previous moment.

[0014] Preferably, the specific steps of obtaining the single-microphone Kalman gain are:

[0015] The covariance of the single Kalman measurement error at the current moment and the covariance of the single Kalman measurement error at the previous moment are obtained, and the single-microphone Kalman gain is calculated based on the covariance of the single Kalman measurement error at the current moment and the covariance of the single Kalman measurement error at the previous moment.

[0016] Preferably, the first-level Kalman adaptive processing includes:

[0017] Obtain a common echo path length range of a first Kalman filter model, obtain a first reference value in the common echo path length range, process the first reference value according to the first Kalman filter model to obtain an array common echo, and subtract the array common echo from the array microphone signal to obtain a first result signal.

[0018] Preferably, the second-level Kalman adaptive processing includes:

[0019] Obtaining a single echo path length range corresponding to the second Kalman filter model, and obtaining a second reference value in the single echo path length range;

[0020] The reverberation filter length corresponding to the second Kalman filter model is obtained according to the second reference value, the second reference value and the reverberation filter length are processed according to the second Kalman filter model to obtain echo reverberation joint information, and the echo reverberation joint information is subtracted from the single microphone signal to obtain a second result signal.

[0021] Preferably, the first reference value is greater than the second reference value.

[0022] Preferably, the echo-reverberation joint information is the sum of the second reference value and the corresponding reverberation filter length.

[0023] In a second aspect, the present application provides a noise cancellation system based on a bipolar Kalman filter, comprising:

[0024] A first acquisition module is used to acquire array microphone signals;

[0025] A second acquisition module is used to obtain a total Kalman gain, and update the total Kalman gain to obtain a first Kalman filter model;

[0026] A first Kalman processing module is used to perform a first-level Kalman adaptive processing on the array microphone signal according to a first Kalman filter model to obtain a first result signal that eliminates the common echo;

[0027] A third acquisition module is used to obtain a single-microphone Kalman gain and update the single-microphone Kalman gain to obtain a second Kalman filter model;

[0028] The second Kalman processing module is used to obtain a single microphone signal according to the first result signal, and perform second-level Kalman adaptive processing on the corresponding single microphone signal according to the second Kalman filter model to obtain a second result signal with residual echo and reverberation eliminated.

[0029] In a third aspect, the present application provides a bipolar Kalman-based noise elimination device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute the bipolar Kalman-based noise elimination method as described above.

[0030] In a fourth aspect, the present application provides a computer-readable storage medium, in which a computer program is stored, wherein the computer program is configured to execute the aforementioned bipolar Kalman-based noise elimination method when running.

[0031] In summary, compared with the prior art, the technical solution provided by this application has at least the following beneficial effects:

[0032] This application obtains array microphone signals, first uses the first Kalman adaptive processing to process the array microphone signals to obtain a first result signal with the common echo removed, and then uses the second-level Kalman adaptive processing to process the single microphone signal in the first result signal to obtain a second result signal with the residual echo and reverberation eliminated. Based on the two-stage Kalman adaptive process, the computational cost of noise elimination can be reduced and the audio quality can be improved without losing the echo filtering speed and depth. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 This is a flow chart of a noise elimination method based on bipolar Kalman as described in an embodiment of the present application.

[0034] Figure 2 This is a schematic diagram of the placement of the array microphone described in an embodiment of the present application.

[0035] Figure 3 This is a module diagram of a bipolar Kalman-based noise elimination system described in an embodiment of the present application.

[0036] Description of reference numerals:

[0037] 1. First acquisition module; 2. Second acquisition module; 3. First Kalman processing module; 4. Third acquisition module; 5. Second Kalman processing module. DETAILED DESCRIPTION

[0038] The following combination Figure 1-Figure 3 To further explain the present application in detail, the terms used in the embodiments of the present application are only for the purpose of describing specific embodiments and are not intended to be limiting.

[0039] Reference Figure 1 The present application relates to a bipolar Kalman-based noise elimination method, which specifically includes:

[0040] Step S1: Acquire array microphone signals;

[0041] Step S2: Obtain the total Kalman gain, and update the total Kalman gain to obtain a first Kalman filter model;

[0042] Step S3: performing a first-level Kalman adaptive processing on the array microphone signal according to the first Kalman filter model to obtain a first result signal with the common echo eliminated;

[0043] Step S4: Obtain a single-microphone Kalman gain, and update a second Kalman filter model according to the single-microphone Kalman gain;

[0044] Step S5: obtaining a single microphone signal according to the first result signal, performing a second-level Kalman adaptive processing on the corresponding single microphone signal according to the second Kalman filter model, and obtaining a second result signal with residual echo and reverberation eliminated.

[0045] Specifically, the present application first acquires array microphone signals and then processes these signals using a first-stage Kalman adaptive process to obtain a first result signal that eliminates the common echo. A second-stage Kalman adaptive process is then used to process the individual microphone signals within the first result signal, further eliminating residual echo and reverberation to obtain a second result signal. The advantage of the two-stage Kalman adaptive process is that it reduces the computational cost of noise cancellation while maintaining the speed and depth of echo filtering, thereby improving audio quality.

[0046] Reference Figure 2, the array microphone signal is composed of multiple microphones, which are arranged together according to a certain geometric structure and can receive sound signals from different directions at the same time. By processing and analyzing these signals, functions such as locating the sound source, noise reduction and sound enhancement are realized. The present application includes but is not limited to the use of eight microphones. The microphones are arranged around the speaker, and each microphone is at the same distance from the speaker. The array formed by the combination of microphones is in the shape of but not limited to a ring. The ring-shaped microphone array can capture sound signals from different directions, making the positioning of the sound source more accurate. By processing and analyzing the signals received by each microphone, the direction and position of the sound source can be determined. The balanced setting of the relative positions and distances between multiple microphones is more conducive to utilizing the mutual relationship between the array signals for noise suppression and noise reduction processing, thereby improving the noise reduction effect.

[0047] Model the signal picked up by the microphone array:

[0048] Y(t)=S(t)+S reverberation (t)+N_echo(t)+N_background(t)

[0049] Among them, Y(t) represents the original signal picked up by the array microphone at time t, that is, the array microphone signal; S(t) is the direct sound transmitted from the target sound source to the array microphone at time t, that is, the sound after removing the noise, S reverberation (t) represents the late reflection of the target sound source transmitted to the array microphone at time t, that is, the reverberation sound; N_echo(t) represents the sound transmitted from the loudspeaker to the array microphone at time t, that is, the echo; N_background(t) represents the background noise picked up by the array microphone at time t. The purpose of the embodiment of the present application is to eliminate the reverberation S in the array microphone signal through a two-stage Kalman adaptive process. reverberation (t) and echo N_echo(t), the background noise N_background can be filtered out by subsequent spatial filter operations such as beamforming.

[0050] Among them, Y(t)=[Y1(t),Y2(t),…,YM(t)]; S(t)=[S1(t),S2(t),…,SM(t)];

[0051] S reverberation (t)=[S reverberation 1(t),S reverberation 2(t),…,S reverberation M(t)];

[0052] N_echo(t)=[N_echo1(t),N_echo2(t),…,N_echoM(t)];

[0053] N_background(t)=[N_background1(t),N_background2(t),…,N_backgroundM(t)];

[0054] M is the total number of microphones in the microphone array.

[0055] As one of the implementation methods, the specific steps for obtaining the total Kalman gain are: obtaining the covariance of the total Kalman measurement error at the current moment and the covariance of the total Kalman measurement error at the previous moment, and calculating the total Kalman gain based on the covariance of the total Kalman error at the current moment and the covariance of the total Kalman error at the previous moment.

[0056] Specifically, the first Kalman filter model update process of the embodiment of the present application is:

[0057]

[0058] In the above formula, K1(t) is the Kalman gain at time t; Φ_s1(t)=E{S(t|t-1)S * (t|t-1)} is the power spectral density of the target sound source and the covariance of the measurement error of the Kalman measurement equation. The superscript * indicates conjugation, and E{} indicates the averaging operation; Φ1(t|t-1) is the covariance of the Kalman process error at the previous time t-1.

[0059] Wkf1(t)=Wkf1(t|t-1)+K1(t)E{S * (t|t-1)}

[0060] The above formula uses the Kalman gain K1(t) to update the Kalman filter.

[0061] Φ1(t)=[I-K1(t)Z1 H (t)]Φ1(t|t-1)

[0062] The above formula uses the Kalman gain K1(t) to update the covariance of the Kalman process error. I is the identity matrix.

[0063] Wkf1(t+1|t)=A(t)Wkf1(t)

[0064] The above formula is the Kalman state equation, which is usually described by a first-order Markov model. A(t) is the state transition parameter at time t and can be taken as a constant of 0.98.

[0065] Φ1(t+1|t)=A(T)Φ1(t)A(T)+Φ_u1(t)

[0066] The above formula uses the Kalman filter after state transfer to update the covariance of the Kalman process error, Φ_u1(t)=Wkf1(t)Wkf1 * (t) is the covariance of the prediction error of the Kalman state equation.

[0067] After the Kalman filter is updated, the covariance of the Kalman process error will act on the next moment t+1.

[0068] As one of the implementation methods, the first-level Kalman adaptive processing includes: obtaining the common echo path length range of the first Kalman filter model, obtaining a first reference value in the common echo path length range, processing the first reference value according to the first Kalman filter model to obtain an array common echo, and subtracting the array common echo from the array microphone signal to obtain a first result signal.

[0069] Specifically, the embodiment of the present application performs a first-level Kalman adaptive processing process. The array adaptively calculates a Kalman filter Wkf1, i.e., the first Kalman filter model, to estimate the common echo path of the array microphones and eliminate the common echo portion in the array microphone signals. Using the Kalman filter, the common echo portion of the array microphones is eliminated as follows, i.e., the Kalman measurement equation:

[0070] S(t|t-1)=Y(t)-Wkf1 H (t|t-1)*Z1(t)

[0071] Where, Y(t) = [Y1(t), Y2(t), …, YM(t)] is the original signal picked up by the array microphone, that is, the array microphone signal, S(t|t-1) = [S1(t|t-1), S2(t|t-1), …, SM(t|t-1)] is the array microphone signal after eliminating the common echo part, Z1(t) = [X(t), …, X(tL aec1 +1)], where X(t) is the audio played by the speaker at time t, L aec1 This is the length of the Kalman filter used in the first-stage Kalman adaptation process, which is also the length of the common echo path for the array microphones. It is typically a large value, such as 400ms, which corresponds to the number of frames. Wkf1(t|t-1) is the Kalman filter estimated at the previous time t-1, denoted by the superscript H. In this embodiment, the first Kalman adaptation process directly uses a single Kalman filter to eliminate the common echo component for the M array microphones.

[0072] Placing the echo cancellation part before beamforming, that is, performing echo cancellation on the original signal of each array microphone, can avoid the influence of beamforming on echo convergence, but its computational cost is very high. Taking multi-channel Kalman echo cancellation as an example, its core algorithm requires ML aecMultiplication and addition operations, usually for echo removal effect, L aec The value of is large, at least 400ms Fourier transform frame number, in order to effectively converge the echo path; however, since the physical position of the array microphone is usually relatively fixed, this application is based on a two-stage Kalman adaptive process, which can effectively reduce the computational cost without losing the speed and depth of echo filtering. First, the first-level Kalman process is used to eliminate the common echo part of the array microphone. Its core algorithm requires L aec1 Multiplication and accumulation operations, L aec The value of can be large, at least 400ms of Fourier transform frames. By taking advantage of the relatively fixed physical position of the array microphones and first eliminating the common echo portion of the array microphones, the computational cost of noise cancellation can be reduced and audio quality can be improved.

[0073] As one implementation method, the specific steps of obtaining the single-microphone Kalman gain are as follows:

[0074] The covariance of the single Kalman measurement error at the current moment and the covariance of the single Kalman measurement error at the previous moment are obtained, and the single-microphone Kalman gain is calculated based on the covariance of the single Kalman measurement error at the current moment and the covariance of the single Kalman measurement error at the previous moment.

[0075] Specifically, the second Kalman filter model update process in the embodiment of the present application is:

[0076]

[0077] The above formula K2m(t) is the Kalman gain of the mth microphone signal at time t; Φ_s2m(t)=S2m(t|t-1)S2m * (t|t-1) is the power spectral density of the target sound source of the m-th microphone signal, and is also the covariance of the measurement error of the Kalman measurement equation of the m-th microphone signal. The superscript * indicates conjugation; Φ2m(t|t-1) is the covariance of the Kalman process error of the m-th microphone signal at the previous time t-1.

[0078] Wkf2m(t)=Wkf2m(t|t-1)+K2m(t)S2m * (t|t-1)

[0079] The above formula uses the Kalman gain K2m(t) to update the Kalman filter of the m-th microphone signal.

[0080] Φ2m(t)=[I-K2m(t)Z2 H (t)]Φ2m(t|t-1)

[0081] The above formula uses the Kalman gain K2m(t) to update the covariance of the Kalman process error of the m-th microphone signal, and I is the unit matrix.

[0082] Wkf2m(t+1|t)=A2m(t)Wkf2m(t)

[0083] The above formula is the Kalman state equation, which is usually described by a first-order Markov. A2m(t) is the state transition parameter at time t and can be taken as a constant of 0.98.

[0084] Φ2m(t+1|t)=A2m(t)Φ2m(t)A2m(t)+Φ_u2m(t)

[0085] The above formula uses the Kalman filter after state transfer to update the covariance of the Kalman process error of the mth microphone signal, Φ_u2m(t)=Wkf2m(t)Wkf2m * (t) is the covariance of the Kalman state equation prediction error for the mth microphone signal. The updated Kalman filter and the covariance of the Kalman process error will be applied to the next time t+1.

[0086] As one of the implementation methods, the second-level Kalman adaptive processing includes: obtaining a single echo path length range corresponding to the second Kalman filter model, and obtaining a second reference value in the single echo path length range; obtaining the reverberation filter length corresponding to the second Kalman filter model according to the second reference value, processing the second reference value and the reverberation filter length according to the second Kalman filter model to obtain echo-reverberation joint information, and subtracting the echo-reverberation joint information from the single microphone signal to obtain a second result signal.

[0087] Specifically, in this embodiment, the array microphone signal that has passed through the first-level Kalman adaptive process is passed to the second-level Kalman adaptive process, where the echo and reverberation are jointly estimated, and M Kalman filters are calculated to perform echo and reverberation removal on each array microphone signal. The subscript m represents the microphone, and its value range is [1, M]. The second-level Kalman adaptive process is used as an example for the second-level Kalman process of the mth microphone signal.

[0088] Using the Kalman filter, the echo and reverberation of the m-th microphone signal are filtered out as follows, that is, the Kalman measurement equation:

[0089] S2m(t|t-1)=Y1m(t)-Wkf2m H (t|t-1)*Z2(t)

[0090] In the above formula, Y1m(t) is the mth microphone signal after filtering out the common echo part through the first-level Kalman adaptive process, that is, Sm(t|t-1), and S2m(t|t-1) is the mth microphone signal after eliminating the residual echo and reverberation.

[0091] Z2(t)=[X(t), ...., X(tL aec2 +1),

[0092] Y11(t-Ldelay),...,Y11(t-Ldelay-Ldr+1)

[0093] ...,

[0094] Y1M(t-Ldelay),....,Y1M(t-Ldelay-Ldr+1)]

[0095] Where Z2(t) is the joint estimate of echo and multi-channel reverberation, X(t) is the audio played by the speaker at time t, and L aec2 is the length of the filter used by the second-level Kalman adaptive process to filter out the echo, and is also the length of the echo path of each microphone in the array. Since the common echo of the array has been eliminated by the first-level Kalman adaptive process, a smaller value can be taken here, such as the number of frames corresponding to 100ms. The first reference value is greater than the second reference value, and L aec2 < <L aec1 , is enough to eliminate the residual echo, which can greatly reduce the amount of calculation. Ldelay is the delay of the input signal, that is, the microphone signal after the first-level Kalman adaptive process, because the reverberation is the delayed part after the signal is reflected. The delayed signal can be used to estimate the reverberation. The echo-reverberation joint information is the sum of the second reference value and the corresponding reverberation filter length. Ldr is the filter length used by the second-level Kalman adaptive to filter out the reverberation. The length of the second-level Kalman filter is L aec2 +MLdr. The second level Kalman process is used to first remove the residual echo of each microphone array. Its core algorithm requires ML aec2 Multiplication and accumulation operations, L aec2 The value of is usually small; so L aec1 +ML aec2 <<ML aec In the second-level Kalman process, echo and reverberation are jointly estimated, effectively improving sound quality. The joint estimation of reverberation can also improve echo filtering in the second-level Kalman process. At the same time, the elimination of reverberation can also improve the effect of subsequent beamforming.

[0096] Reference Figure 3 , an embodiment of the present application provides a noise cancellation system based on a bipolar Kalman filter, the system comprising:

[0097] A first acquisition module 1 is used to acquire array microphone signals;

[0098] A second acquisition module 2 is used to obtain a total Kalman gain and update the total Kalman gain to obtain a first Kalman filter model;

[0099] A first Kalman processing module 3 is configured to perform a first-level Kalman adaptive processing on the array microphone signal according to a first Kalman filter model to obtain a first result signal that eliminates the common echo;

[0100] A third acquisition module 4 is used to obtain a single-microphone Kalman gain, and update the single-microphone Kalman gain to obtain a second Kalman filter model;

[0101] The second Kalman processing module 5 is used to obtain a single microphone signal according to the first result signal, and perform a second-level Kalman adaptive processing on the corresponding single microphone signal according to the second Kalman filter model to obtain a second result signal with residual echo and reverberation eliminated.

[0102] An embodiment of the present application provides a bipolar Kalman-based noise elimination device, comprising a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute the bipolar Kalman-based noise elimination method as described above.

[0103] An embodiment of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, wherein the computer program is configured to execute the aforementioned bipolar Kalman-based noise elimination method when running.

[0104] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described devices and products can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0105] In several embodiments provided in this application, it should be understood that the disclosed methods, systems, devices, and program products may be implemented in other ways.

[0106] In addition, the functional units in the various embodiments of the present application may be integrated into a single processing unit, or each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0107] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the above embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the above embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A noise elimination method based on bipolar Kalman, characterized in that: include: Get array microphone signal; Obtaining a total Kalman gain, and updating the total Kalman gain to obtain a first Kalman filter model; Performing first-level Kalman adaptive processing on the array microphone signal according to the first Kalman filter model to obtain a first result signal that eliminates the common echo; wherein the first-level Kalman adaptive processing includes: obtaining a common echo path length range of the first Kalman filter model, obtaining a first reference value within the common echo path length range, processing the first reference value according to the first Kalman filter model to obtain an array common echo, and subtracting the array common echo from the array microphone signal to obtain the first result signal; Obtaining a single-microphone Kalman gain, and updating a second Kalman filter model according to the single-microphone Kalman gain; A single microphone signal is obtained based on the first result signal, and the corresponding single microphone signal is subjected to second-level Kalman adaptive processing according to the second Kalman filter model to obtain a second result signal with residual echo and reverberation eliminated, wherein the second-level Kalman adaptive processing includes: obtaining a single echo path length range corresponding to the second Kalman filter model, and obtaining a second reference value in the single echo path length range; obtaining a reverberation filter length corresponding to the second Kalman filter model according to the second reference value, processing the second reference value and the reverberation filter length according to the second Kalman filter model to obtain echo-reverberation joint information, and subtracting the echo-reverberation joint information from the single microphone signal to obtain a second result signal.

2. The noise elimination method based on bipolar Kalman filter according to claim 1, characterized in that: The specific steps of obtaining the total Kalman gain are: The covariance of the total Kalman measurement error at the current moment and the covariance of the total Kalman measurement error at the previous moment are obtained, and the total Kalman gain is calculated according to the covariance of the total Kalman error at the current moment and the covariance of the total Kalman error at the previous moment.

3. The noise elimination method based on bipolar Kalman filter according to claim 1, characterized in that: The specific steps of obtaining the single-microphone Kalman gain are: The covariance of the single Kalman measurement error at the current moment and the covariance of the single Kalman measurement error at the previous moment are obtained, and the single-microphone Kalman gain is calculated based on the covariance of the single Kalman measurement error at the current moment and the covariance of the single Kalman measurement error at the previous moment.

4. The noise elimination method based on bipolar Kalman filter according to claim 1, characterized in that: The first reference value is greater than the second reference value.

5. The noise elimination method based on bipolar Kalman filter according to claim 4, characterized in that: The echo-reverberation joint information is the sum of the second reference value and the corresponding reverberation filter length.

6. A noise cancellation system based on bipolar Kalman, characterized in that: include: A first acquisition module is used to acquire array microphone signals; A second acquisition module is used to obtain a total Kalman gain, and update the total Kalman gain to obtain a first Kalman filter model; a first Kalman processing module, configured to perform first-level Kalman adaptive processing on the array microphone signal according to a first Kalman filter model to obtain a first result signal that eliminates the common echo; specifically, configured to obtain a common echo path length range of the first Kalman filter model, obtain a first reference value within the common echo path length range, process the first reference value according to the first Kalman filter model to obtain an array common echo, and subtract the array common echo from the array microphone signal to obtain a first result signal; A third acquisition module is used to obtain a single-microphone Kalman gain and update the single-microphone Kalman gain to obtain a second Kalman filter model; The second Kalman processing module is used to obtain a single microphone signal based on the first result signal, and perform second-level Kalman adaptive processing on the corresponding single microphone signal according to the second Kalman filter model to obtain a second result signal that eliminates residual echo and reverberation; it is specifically used to obtain the single echo path length range corresponding to the second Kalman filter model, and obtain a second reference value in the single echo path length range; obtain the reverberation filter length corresponding to the second Kalman filter model according to the second reference value, process the second reference value and the reverberation filter length according to the second Kalman filter model to obtain echo-reverberation joint information, and subtract the echo-reverberation joint information from the single microphone signal to obtain the second result signal.

7. A noise elimination device based on bipolar Kalman, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to perform the bipolar Kalman-based noise cancellation method according to any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, wherein the computer program is configured to execute the bipolar Kalman-based noise elimination method according to any one of claims 1 to 5 when running.

Images