CN108172231A - A kind of dereverberation method and system based on Kalman filtering - Google Patents
A kind of dereverberation method and system based on Kalman filtering Download PDFInfo
- Publication number
- CN108172231A CN108172231A CN201711285885.4A CN201711285885A CN108172231A CN 108172231 A CN108172231 A CN 108172231A CN 201711285885 A CN201711285885 A CN 201711285885A CN 108172231 A CN108172231 A CN 108172231A
- Authority
- CN
- China
- Prior art keywords
- signal
- microphone
- kalman
- formula
- variance
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
- G10L21/00—Processing of the speech or voice signal to produce another audible or non-audible signal, e.g. visual or tactile, in order to modify its quality or its intelligibility
- G10L21/02—Speech enhancement, e.g. noise reduction or echo cancellation
- G10L21/0208—Noise filtering
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
- G10L19/00—Speech or audio signals analysis-synthesis techniques for redundancy reduction, e.g. in vocoders; Coding or decoding of speech or audio signals, using source filter models or psychoacoustic analysis
- G10L19/04—Speech or audio signals analysis-synthesis techniques for redundancy reduction, e.g. in vocoders; Coding or decoding of speech or audio signals, using source filter models or psychoacoustic analysis using predictive techniques
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
- G10L19/00—Speech or audio signals analysis-synthesis techniques for redundancy reduction, e.g. in vocoders; Coding or decoding of speech or audio signals, using source filter models or psychoacoustic analysis
- G10L19/04—Speech or audio signals analysis-synthesis techniques for redundancy reduction, e.g. in vocoders; Coding or decoding of speech or audio signals, using source filter models or psychoacoustic analysis using predictive techniques
- G10L19/26—Pre-filtering or post-filtering
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
- G10L21/00—Processing of the speech or voice signal to produce another audible or non-audible signal, e.g. visual or tactile, in order to modify its quality or its intelligibility
- G10L21/02—Speech enhancement, e.g. noise reduction or echo cancellation
- G10L21/0208—Noise filtering
- G10L21/0216—Noise filtering characterised by the method used for estimating noise
- G10L21/0232—Processing in the frequency domain
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
- G10L21/00—Processing of the speech or voice signal to produce another audible or non-audible signal, e.g. visual or tactile, in order to modify its quality or its intelligibility
- G10L21/02—Speech enhancement, e.g. noise reduction or echo cancellation
- G10L21/0208—Noise filtering
- G10L2021/02082—Noise filtering the noise being echo, reverberation of the speech
Abstract
The invention discloses a kind of dereverberation method and system based on Kalman filtering, the method includes：The collected original signal of each microphone is pre-processed to obtain corresponding frequency-region signal, input signal is formed after delay；Using Kalman filtering algorithm and the multichannel autoregression model of time-varying estimation reverb signal, using the collected original signal of each microphone at current time as with reference to signal, subtract reverb signal and obtain error signal；Utilize the coefficient of kalman gain matrix and error signal update Kalman filter；Echo signal is obtained using the collected original signal of current time each microphone, input signal and updated Kalman filter coefficient；Finally, frequency domain echo signal is transformed into time domain using inverse Fourier transform.The method of the present invention reduces the complexity of self-adapting multi-channel linear prediction dereverberation algorithm by diagonalization kalman filter state vector error covariance matrix.
Description
Technical field
The present invention relates to speech dereverbcration field, more particularly to a kind of dereverberation method based on Kalman filtering and it is
System.
Background technology
As shown in Figure 1, since object is to the reflex of sound wave in room boundaries and room, microphone is except receiving sound source
Outside the direct sound wave sent out, also from the reflected sound of all directions.Generally by the sound of arrival time 30-50ms after direct sound wave
Signal is known as reflection, and the acoustical signal reached after this is known as late period reflected sound, i.e. reverberation is trailed.Psychologic acoustics research
It was found that reflection can enhance the intensity of direct sound wave, the intelligibility of speech is improved.And reverb signal can shelter subsequently reach it is straight
Up to acoustical signal, lead to voice fuzzy.In addition, reverb signal can also reduce the voice quality and voice that microphone receives signal
Identifying system accurately identifies rate.Under the application scenarios such as the videoconference, the intelligent sound box that are carried out in closed room, microphone is past
Toward the far field for being in sound source.With the increase of distance between sound source and microphone, the destruction that reverberation receives microphone signal is made
With more seriously.In addition, in voice communication system, ambient noise is smaller, and the signal that microphone receives mainly is mixed by room
Loud influence causes voice signal accuracy and intelligibility all to be declined, seriously affects communication quality.Therefore, to microphone
It is a very necessary job to receive signal dereverberation.
Speech dereverbcration is a popular research topic.Current solution mainly has：
(1) linear predictive residual enhancing algorithm.The speech model that linear predictive residual enhancing algorithm utilizes is filtered for sound source
Device model.Voice is regarded as all-pole filter of a string of activation sequences by a time-varying in the model.Reverberation voice is believed
It number can obtain the estimated value of all-pole filter coefficient, that is, linear predictor coefficient as linear prediction analysis.Then to Mike
Wind receives signal and makees liftering, you can obtains corresponding pumping signal, that is, residual signals.By enhancing residual signals
Realize dereverberation, it can reconstructed speech signal by the linear predictor coefficient for estimating to obtain.
(2) Enhancement Method is composed.Spectrum Enhancement Method is a kind of classical dereverberation algorithm.This method passes through in Fourier in short-term
Transform domain amendment is noisy or containing reverb signal, achievees the purpose that enhance voice signal.Document [1] (K.Kinoshita,
M.Delcroix,T.Nakatani,and M.Miyoshi,“Suppression of late reverberation
effecton speech signal using long-term multiple-step linear prediction,”IEEE
Trans.Audio, Speech, Lang.Process., vol.17, no.4, pp.534-545, May 2009.) pass through delay line
Property predictive estimation late reverberation, subsequent spectrum-subtraction is recycled to realize dereverberation.Document [2] (F.Xiong, N.Moritz,
R.Rehr,J.Anemuller,B.Meyer,T.G.G.Doclo,and S.Goetze,“Robust ASR in
reverberant environments using temporal cepstrum smoothing for speech
enhancement and an amplitude modulation filterbank for feature extraction,”in
Proc.REVERB Challenge Workshop, Florence, Italy, 2014.) estimated using minimum mean square error method
Clean voice signal amplitude spectrum, as the pretreatment stage of automatic speech recognition, by late reverberation and steady ambient noise
Power spectral density can estimate the power spectral density of clean speech signal.Under normal circumstances, spectrum Enhancement Method is determines spectrum attenuation etc.
Grade needs first to estimate the reverberation time.However, the problem of blind reverberation estimation is still very difficult, especially in noise-containing ring
Border, the research of the problem is still in continuous progress.
(3) liftering method.Blind dereverberation algorithm refers to during dereverberation, the room between sound source and microphone
Between the priori of impulse response be unknown.Multichannel linear prediction algorithm based on microphone array is a kind of the blind of classics
Dereverberation algorithm.According to Multiinputoutput invert it is theoretical (Multiple input/output inverse theorem,
MINT), under conditions of each channel transfer function is without common zero point, multi-channel method room constant when can be with perfect equilibrium
Impulse response.However, MINT algorithms are very sensitive to System Discrimination error, and the impulse response of actual room often contains phase
Near zero, therefore MINT algorithms are difficult to apply in practice.
Since time domain linear prediction algorithm often requires that very long filter length, and asking there are albefaction echo signal
Topic.There is scholar to propose using multichannel linear prediction algorithm in each subband independent process to believe in Short Time Fourier Transform domain recently
Number.In STFT domains, reverberation voice signal is described in each frequency band with autoregression model, it is possible thereby to reduce the filtering of each subband
Device length.Since room impulse response is actually what is changed over time, so the prediction model coefficient of time-varying is needed to model.Most
Closely there is multichannel autoregression (Multichannel autoregressive, MAR) signal model that scholar proposes STFT domains,
Estimate MAR coefficients using Kalman filter, which can be considered a kind of recurrence least square (Recursiveleast of broad sense
Squares, RLS) algorithm.
The computation complexity of multichannel linear prediction algorithm based on STFT domains and each sub-filter exponent number into square
Relationship.The complexity limits application of the algorithm in the limited system platform of many resources.Document [3] (Dietzen T,
Doclo S,Spriet A,et al.Low-complexity Kalmanfilterformulti-channel linear-
prediction-basedblindspeechdereverberarion[C].IEEE Workshop on Applications
Of Signal Processing to Audio and Acoustics.IEEE, 2017.) for the adaptive mostly logical of STFT domains
Road linear prediction dereverberation algorithm, it is proposed that computation complexity is dropped to and filtered by a kind of Kalman filtering method for solving of simplification
Wave device exponent number is linear.However, the simplification method can cause a degree of voice quality to decline.In addition, the algorithm is only
Estimate a channel signal, need to calculate multiple channels in practice.
Invention content
It is an object of the invention to overcome drawbacks described above existing for current dereverberation method, propose a kind of based on Kalman's filter
It is adaptive further to reduce STFT domains while ensureing not lose voice quality for the low complex degree dereverberation method of wave, this method
Answer the complexity of multichannel linear prediction dereverberation algorithm.
For achieving the above object, the present invention proposes a kind of dereverberation method based on Kalman filtering, this method packet
It includes：
The collected original signal of each microphone is pre-processed to obtain corresponding frequency-region signal, input is formed after delay
Signal；
It is each by current time using Kalman filtering algorithm and the multichannel autoregression model of time-varying estimation reverb signal
The collected original signal of microphone is used as with reference to signal, is subtracted reverb signal and is obtained error signal；
Utilize the coefficient of kalman gain matrix and error signal update Kalman filter；
Utilize the collected original signal of current time each microphone, input signal and updated Kalman filter system
Number obtains echo signal；
Finally, frequency domain echo signal is transformed into time domain using inverse Fourier transform.
As a kind of improvement of the above method, the method specifically includes：
Step 1) is by the collected signal y of M microphonem(n), 1≤m≤M progress framing, adding window and Fourier transformation obtains
To corresponding frequency-region signal Ym(n),
Frequency-region signal Ym(n) it is：
Wherein, k is frequency index, and N is the points of Fourier transformation；N be time frame subscript, wSTFT(l) in Fu in short-term
Leaf transformation analyzes window function, and R represents frame shifting；
Step 2) forms input signal matrix Y (n-D) by the frequency-region signal of the M microphone at n-D to n-L moment, utilizes
Kalman's weight vectors estimation reverb signal vector r (n), wherein D are delay, and L is linear prediction length；
Y (n)=[Y1(n),...,YM(n)]T (2)
In formula (3), IMIt is the unit matrix of M × M,Kronecker products are represented, Y (n-D) is by microphone observation signal
The size of composition is M × LcSparse matrix, Lc=M2(L-D+1)；
Reverb signal vector r (n) is calculated according to formula (4)；
In formula (4),The Matrix C of M × Mp(n-1) it is time-varying
Kalman's weight vectors coefficient, p=[D, D+1 ..., L], Vec { } are the rectangular array stack operation factor；
The reverberation that step 3) subtracts the step 2) acquisition using the signal y (n) of current time each microphone acquisition is believed
Number vector r (n) obtains error signal vector e (n)；
E (n)=y (n)-r (n) (5)
Step 4) calculates kalman gain matrix K (n)；
Step 5) updates Kalman filter coefficient by kalman gain matrix K (n) and error signal vector e (n)
Signal y (n), input signal matrix Y (n-D) and the updated card that step 6) is acquired using current time microphone
Thalmann filter coefficientCalculate echo signal vector x (n)；
Step 7) carries out inverse Fourier transform to frequency domain echo signal vector x (n), obtains time domain echo signal vector xt
(l)：
As a kind of improvement of the above method, the step 4) specifically includes：
Step 401) is calculated according to formula (6) in a manner that single order is smooth
Wherein,For the echo signal variance at n-1 moment,For the echo signal variance at n-2 moment, x
(n-1) it is n-1 moment echo signal vector；α is smoothing factor, value 0.2；
Step 402) is according to formula (7) variance of calculation perturbation noise w (n) firstThen it is calculated first according to formula (8)
Test imbalance variance
In formula (7), Lc=M2(L-D+1), η is usually 10-5；Posteriority imbalance variance for the n-1 moment；
Step 403) is according to formula (9) by echo signal varianceWith priori imbalance varianceCalculate the Regularization factor
δ(n)；
Step 404) calculates covariance matrix S according to formula (10) by the collected signal of microphoneY(n-D)；
SY(n-D)=Y (n-D) YH(n-D) (10)
Step 405) calculates kalman gain matrix K (n) according to formula (11)；
K (n)=YH(n-D)[SY(n-D)+δ(n)IM]-1 (11)。
As a kind of improvement of the above method, further included after the step 7)：
Update posteriority imbalance variance
A kind of dereverberation system based on Kalman filtering, including memory, processor and storage on a memory and
The computer program that can be run on a processor, which is characterized in that the processor realizes the above method when performing described program
The step of.
The advantage of the invention is that：
1st, method of the invention is reduced adaptive by diagonalization kalman filter state vector error covariance matrix
Answer the complexity of multichannel linear prediction dereverberation algorithm；
2nd, it is minimum to be considered as a kind of normalization for becoming the Regularization factor for the Kalman filtering algorithm of simplification of the invention
Side's (Normalized Least Mean Square, NLMS) algorithm.In addition, the Kalman filtering of simplification proposed by the present invention is calculated
The error signal vector e (n) and echo signal vector x (n) of method are the vector of M × 1, and for subsequent cascaded, other multichannels are calculated for this
Method provides a convenient.In addition, also it is the variance for calculating echo signalProvide more available informations.
Description of the drawings
Fig. 1 generates schematic diagram for RMR room reverb；
Fig. 2 is the block diagram of the Kalman filtering dereverberation of the present invention；
Fig. 3 is the newer block diagram of Kalman's weight vector of the present invention；
Fig. 4 is the block diagram of the calculating kalman gain matrix module of the present invention；
Fig. 5 is the block diagram of the estimation priori imbalance variance of the present invention.
Specific embodiment
The present invention will be described in detail in the following with reference to the drawings and specific embodiments.
A kind of low complex degree dereverberation method based on Kalman filtering, the method includes：
Step 1) is by the collected signal y of M microphonem(n), 1≤m≤M progress framing, adding window and Fourier transformation obtains
To corresponding frequency-region signal Ym(k, n) is represented to simplify, and will hereinafter omit frequency index k；
Frequency-region signal YmThe calculating of (k, n) is calculated according to formula (1)：
Wherein, k is frequency index, and N is the points of Fourier transformation；N be time frame subscript, wSTFT(l) in Fu in short-term
Leaf transformation analyzes window function, and R represents frame shifting；
Step 2) forms input signal matrix Y (n-D) by the frequency-region signal of the M microphone at n-D to n-L moment, utilizes
Kalman's weight vectors estimation reverb signal vector r (n), wherein D are delay, and L is linear prediction length；
Y (n-D) is that the size being made of microphone observation signal is M × LcSparse matrix, Lc=M2(L-D+1)。r(n)
Represent late reverberation.
Input signal matrix Y (n-D) is obtained according to formula (2) and (3)；
Y (k, n)=[Y1(k,n),...,YM(k,n)]T (2)
In formula (3),Represent Kronecker products.
Reverb signal vector r (n) is calculated according to formula (4)；
In formula (4),Represent the estimated value to a certain signal,M
The Matrix C of × Mp(n-1) it is Kalman's weight vectors coefficient coefficient of time-varying, p=[D, D+1 ..., L].L is long for linear prediction
Degree, the selection of delay D ＞ 1 is related with the frame Overlapping parameters of STFT (Short-time Fourier transform, STFT), takes
Value, which will ensure that x (n) is related to r's (n), to be ignored.Vec { } is the rectangular array stack operation factor.
The reverberation that step 3) subtracts the step 2) acquisition using the signal y (n) of current time each microphone acquisition is believed
Number vector r (n) obtains error signal vector e (n)；
E (n)=y (n)-r (n) (5)
Step 4) is by input signal matrix Y (n-D), echo signal varianceWith priori imbalance varianceCalculate card
Germania gain matrix K (n)；It specifically includes：
Step 401) calculates the echo signal variance at n moment according to formula (6) in a manner that single order is smooth
Wherein,For the echo signal variance at n-1 moment,For the echo signal variance at n-2 moment, x
(n-1) it is n-1 moment echo signal vector；α is smoothing factor, value 0.2；
Step 402) is according to formula (7) variance of calculation perturbation noise w (n) firstThen it is calculated first according to formula (8)
Test imbalance variance
In formula (7), Lc=M2(L-D+1), η is a small normal number, and general recommendations takes 10-5。
Step 403) is according to formula (9) by echo signal varianceWith priori imbalance varianceCalculate the Regularization factor
δ(n)；
Step 404) calculates covariance matrix S according to formula (10) by the collected signal of microphoneY(n-D)；
SY(n-D)=Y (n-D) YH(n-D) (10)
Step 405) calculates kalman gain matrix K (n) according to formula (11)；
K (n)=YH(n-D)[SY(n-D)+δ(n)IM]-1 (11)
Step 5) updates Kalman filter coefficient by kalman gain matrix K (n) and error signal vector e (n)
Signal y (n), input signal matrix Y (n-D) and the updated card that step 6) is acquired using current time microphone
Thalmann filter coefficientCalculate echo signal vector x (n)；
Step 7) asks the inverse Fourier transform of frequency-region signal vector x (n), obtains time domain echo signal vector xt(l)；
Step 8) update posteriority imbalance variance
In formula (15), IMIt is the unit matrix of M × M, Lc=M2(L-D+1), L is linear prediction length.Square is sought in tr [] expressions
The mark of battle array.
As shown in Fig. 2, Fig. 2 is the low complex degree dereverberation algorithmic system block diagram based on Kalman filtering of the present invention.Its
In, Y (n-D) is the input signal matrix being made of the frequency-region signal of the M microphone at n-D to n-L moment, and r (n) is by karr
The reverb signal vector that graceful filtering algorithm estimates, y (n) is the reference signal that the signal acquired by current time microphone is formed
Vector, x (n) are the echo signal vector of final output.Fourier transformation module 201 represents to carry out the signal that microphone acquires
Fourier transformation, the Fourier transformation Y of m-th of microphone signalm(n) it represents.Time delay module 202 represents to acquire microphone
Signal carry out delay operation.The selection for postponing D ＞ 1 is related with the frame Overlapping parameters of STFT, and value will ensure x (n) and r (n)
Correlation can ignore.Kalman filtering module 203 represents to be filtered input signal using Kalman filter, and estimation is mixed
Ring signal.Echo signal vector x (n) is calculated by summation module 204.Inverse Fourier transform module 205 becomes frequency-region signal
Change to time domain.
Fig. 3 updates functional block diagram for Kalman's weight coefficient, wherein including kalman gain computing module 303.Believed by error
Number vector, kalman gain matrix obtain the renewal amount of weight vector, and the mesh of final output can be calculated by newer weight vector
Mark signal vector x (n).
Fig. 4 is the functional block diagram for calculating kalman gain matrix, wherein including priori imbalance variance evaluation module 403.Multiply
Volume module 401 realizes that two input variables are multiplied, and module of inverting 402 represents that input signal is carried out to take inverse operation.Believed using target
Number varianceInput signal matrix Y (n-D) and priori offset errorCalculate kalman gain matrix.
It is calculated by priori imbalance variance evaluation module 403.Kalman gain lacks of proper care to the update of wave filter weight coefficient and priori
The estimation of variance is most important.R is calculated firste(n), kalman gain matrix K (n) is then calculated.
Priori imbalance variance evaluation module shown in fig. 5 also reflects posteriority imbalance varianceComputational methods.Transposition
Module 501 represents to carry out transposition operation to matrix.Module 503 represents to seek the mark of matrix.
It can obtain to draw a conclusion by above-mentioned analysis and Fig. 2, Fig. 3 and Fig. 4：
First, after using the technology of the present invention, STFT domains self-adapting multi-channel linear prediction dereverberation algorithm is greatly reduced
Computation complexity；
Secondly, after the technology of the present invention, computation complexity is not only reduced, the voice quality of output is also protected
Card；
Finally, it after using the technology of the present invention, can be obtained between the tracking performance and constringency performance of Kalman filter
Compromise well.
Adequately show above the present invention provides a kind of effective dereverberation technology, can be very good removal due to room
Reverberation interference caused by sound reflecting, the raising intelligibility of speech and automatic speech recognition system accurately identify rate.
It should be noted that the Kalman filtering algorithm of simplification described in the invention be considered as it is a kind of change Regularization because
The NLMS algorithms of son, wherein δ (n) can be considered a variable Regularization factor.VarianceFilter coefficient c (n) is estimated
Meter plays an important roll, smallerValue characterizes the tracking performance of good imbalance performance and difference, largerValue
Characterize the imbalance performance of good tracking performance and difference.In other words,Value highly determine Kalman filter
Tracking performance and constringency performance.When algorithm is also not converged,WithDifference it is larger, according to formula (7),
Larger value is also taken at this time, thus provides quick constringency performance and tracking performance.When algorithm starts to converge to stable state,WithDifference reduce, result in smallerNamely relatively low imbalance.
It should be noted last that the above embodiments are merely illustrative of the technical solutions of the present invention and it is unrestricted.Although ginseng
The present invention is described in detail according to embodiment, it will be understood by those of ordinary skill in the art that, to the technical side of the present invention
Case is modified or replaced equivalently, and without departure from the spirit and scope of technical solution of the present invention, should all be covered in the present invention
Right in.
Claims (5)
1. a kind of dereverberation method based on Kalman filtering, the method includes：
The collected original signal of each microphone is pre-processed to obtain corresponding frequency-region signal, input letter is formed after delay
Number；
Using Kalman filtering algorithm and the multichannel autoregression model of time-varying estimation reverb signal, by each Mike at current time
The collected original signal of wind is used as with reference to signal, is subtracted reverb signal and is obtained error signal；
Utilize the coefficient of kalman gain matrix and error signal update Kalman filter；
It is obtained using the collected original signal of current time each microphone, input signal and updated Kalman filter coefficient
To echo signal；
Finally, frequency domain echo signal is transformed into time domain using inverse Fourier transform.
2. the dereverberation method according to claim 1 based on Kalman filtering, which is characterized in that the method is specifically wrapped
It includes：
Step 1) is by the collected signal y of M microphonem(n), 1≤m≤M carries out framing, adding window and Fourier transformation and obtains phase
The frequency-region signal Y answeredm(n),
Frequency-region signal Ym(n) it is：
Wherein, k is frequency index, and N is the points of Fourier transformation；N be time frame subscript, wSTFT(l) it is Short Time Fourier Transform
Window function is analyzed, R represents frame shifting；
Step 2) forms input signal matrix Y (n-D) by the frequency-region signal of the M microphone at n-D to n-L moment, utilizes karr
Graceful weight vectors estimation reverb signal vector r (n), wherein D are delay, and L is linear prediction length；
Y (n)=[Y1(n),...,YM(n)]T (2)
In formula (3), IMIt is the unit matrix of M × M,Kronecker products are represented, Y (n-D) is made of microphone observation signal
Size be M × LcSparse matrix, Lc=M2(L-D+1)；
Reverb signal vector r (n) is calculated according to formula (4)；
In formula (4),The Matrix C of M × Mp(n-1) it is the karr of time-varying
Graceful weight vectors coefficient, p=[D, D+1 ..., L], Vec { } are the rectangular array stack operation factor；
Step 3) using the signal y (n) of current time each microphone acquisition subtract reverb signal that the step 2) obtains to
Amount r (n) obtains error signal vector e (n)；
E (n)=y (n)-r (n) (5)
Step 4) calculates kalman gain matrix K (n)；
Step 5) updates Kalman filter coefficient by kalman gain matrix K (n) and error signal vector e (n)
Signal y (n), input signal matrix Y (n-D) and the updated Kalman that step 6) is acquired using current time microphone
Filter coefficientCalculate echo signal vector x (n)；
Step 7) carries out inverse Fourier transform to frequency domain echo signal vector x (n), obtains time domain echo signal vector xt(l)：
3. the dereverberation method according to claim 2 based on Kalman filtering, which is characterized in that the step 4) is specific
Including：
Step 401) is calculated according to formula (6) in a manner that single order is smooth
Wherein,For the echo signal variance at n-1 moment,For the echo signal variance at n-2 moment, x (n-
1) it is n-1 moment echo signal vector；α is smoothing factor, value 0.2；
Step 402) is according to formula (7) variance of calculation perturbation noise w (n) firstThen priori is calculated according to formula (8) to lose
Adjust variance
In formula (7), Lc=M2(L-D+1), η is usually 10-5；Posteriority imbalance variance for the n-1 moment；
Step 403) is according to formula (9) by echo signal varianceWith priori imbalance varianceCalculate Regularization factor delta
(n)；
Step 404) calculates covariance matrix S according to formula (10) by the collected signal of microphoneY(n-D)；
SY(n-D)=Y (n-D) YH(n-D) (10)
Step 405) calculates kalman gain matrix K (n) according to formula (11)；
K (n)=YH(n-D)[SY(n-D)+δ(n)IM]-1 (11)。
4. the dereverberation method according to claim 3 based on Kalman filtering, which is characterized in that the step 7) is gone back afterwards
Including：
Update posteriority imbalance variance
5. a kind of dereverberation system based on Kalman filtering, including memory, processor and storage on a memory and can
The computer program run on a processor, which is characterized in that the processor realized when performing described program claim 1~
The step of one of 4 the method.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711285885.4A CN108172231B (en) | 2017-12-07 | 2017-12-07 | Dereverberation method and system based on Kalman filtering |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711285885.4A CN108172231B (en) | 2017-12-07 | 2017-12-07 | Dereverberation method and system based on Kalman filtering |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108172231A true CN108172231A (en) | 2018-06-15 |
| CN108172231B CN108172231B (en) | 2021-07-30 |
Family
ID=62524587
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711285885.4A Active CN108172231B (en) | 2017-12-07 | 2017-12-07 | Dereverberation method and system based on Kalman filtering |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108172231B (en) |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108600894A (en) * | 2018-07-11 | 2018-09-28 | 重庆传乐音响科技有限公司 | A kind of earphone adaptive active noise control system and method |
| CN109297718A (en) * | 2018-09-29 | 2019-02-01 | 重庆长安汽车股份有限公司 | A kind of evaluation method of order whistler |
| CN110289011A (en) * | 2019-07-18 | 2019-09-27 | 大连理工大学 | A kind of speech-enhancement system for distributed wireless acoustic sensor network |
| WO2020078210A1 (en) * | 2018-10-18 | 2020-04-23 | 电信科学技术研究院有限公司 | Adaptive estimation method and device for post-reverberation power spectrum in reverberation speech signal |
| CN111474481A (en) * | 2020-04-13 | 2020-07-31 | 深圳埃瑞斯瓦特新能源有限公司 | Battery SOC estimation method and device based on extended Kalman filtering algorithm |
| CN111540372A (en) * | 2020-04-28 | 2020-08-14 | 北京声智科技有限公司 | Method and device for multi-microphone array noise reduction processing |
| CN111599372A (en) * | 2020-04-02 | 2020-08-28 | 云知声智能科技股份有限公司 | Stable on-line multi-channel voice dereverberation method and system |
| CN111599374A (en) * | 2020-04-16 | 2020-08-28 | 云知声智能科技股份有限公司 | Single-channel voice dereverberation method and device |
| CN111933170A (en) * | 2020-07-20 | 2020-11-13 | 歌尔科技有限公司 | Voice signal processing method, device, equipment and storage medium |
| CN114205731A (en) * | 2021-12-08 | 2022-03-18 | 随锐科技集团股份有限公司 | Speaker area detection method, device, electronic equipment and storage medium |
| CN115065422A (en) * | 2021-07-26 | 2022-09-16 | 中国计量科学研究院 | System and method for evaluating communication quality in reverberant room |
| CN117316175A (en) * | 2023-11-28 | 2023-12-29 | 山东放牛班动漫有限公司 | Intelligent encoding storage method and system for cartoon data |
| CN117318671A (en) * | 2023-11-29 | 2023-12-29 | 有研(广东)新材料技术研究院 | Self-adaptive filtering method based on fast Fourier transform |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101460999A (en) * | 2006-06-05 | 2009-06-17 | 埃克奥迪公司 | Blind signal extraction |
| CN103187068A (en) * | 2011-12-30 | 2013-07-03 | 联芯科技有限公司 | Priori signal-to-noise ratio estimation method, device and noise inhibition method based on Kalman |
| US20130332156A1 (en) * | 2012-06-11 | 2013-12-12 | Apple Inc. | Sensor Fusion to Improve Speech/Audio Processing in a Mobile Device |
| CN107393550A (en) * | 2017-07-14 | 2017-11-24 | 深圳永顺智信息科技有限公司 | Method of speech processing and device |
-
2017
- 2017-12-07 CN CN201711285885.4A patent/CN108172231B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101460999A (en) * | 2006-06-05 | 2009-06-17 | 埃克奥迪公司 | Blind signal extraction |
| CN103187068A (en) * | 2011-12-30 | 2013-07-03 | 联芯科技有限公司 | Priori signal-to-noise ratio estimation method, device and noise inhibition method based on Kalman |
| US20130332156A1 (en) * | 2012-06-11 | 2013-12-12 | Apple Inc. | Sensor Fusion to Improve Speech/Audio Processing in a Mobile Device |
| CN107393550A (en) * | 2017-07-14 | 2017-11-24 | 深圳永顺智信息科技有限公司 | Method of speech processing and device |
Non-Patent Citations (2)
| Title |
|---|
| CHRISTINE EVERS ET AL.: "Multichannel Online Blind Speech Dereverberation with Marginalization of Static Observation Parameters in a Rao-Blackwellized Particle Filter", 《JOURNAL OF SIGNAL PROCESSING SYSTEMS》 * |
| SEBASTIAN BRAUN ET AL.: "Online Dereverberation for Dynamic Scenarios Using a Kalman Filter With an Autoregressive Model", 《IEEE SIGNAL PROCESSING LETTERS》 * |
Cited By (20)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108600894A (en) * | 2018-07-11 | 2018-09-28 | 重庆传乐音响科技有限公司 | A kind of earphone adaptive active noise control system and method |
| CN109297718A (en) * | 2018-09-29 | 2019-02-01 | 重庆长安汽车股份有限公司 | A kind of evaluation method of order whistler |
| WO2020078210A1 (en) * | 2018-10-18 | 2020-04-23 | 电信科学技术研究院有限公司 | Adaptive estimation method and device for post-reverberation power spectrum in reverberation speech signal |
| CN110289011B (en) * | 2019-07-18 | 2021-06-25 | 大连理工大学 | Voice enhancement system for distributed wireless acoustic sensor network |
| CN110289011A (en) * | 2019-07-18 | 2019-09-27 | 大连理工大学 | A kind of speech-enhancement system for distributed wireless acoustic sensor network |
| CN111599372A (en) * | 2020-04-02 | 2020-08-28 | 云知声智能科技股份有限公司 | Stable on-line multi-channel voice dereverberation method and system |
| CN111599372B (en) * | 2020-04-02 | 2023-03-21 | 云知声智能科技股份有限公司 | Stable on-line multi-channel voice dereverberation method and system |
| CN111474481A (en) * | 2020-04-13 | 2020-07-31 | 深圳埃瑞斯瓦特新能源有限公司 | Battery SOC estimation method and device based on extended Kalman filtering algorithm |
| CN111599374A (en) * | 2020-04-16 | 2020-08-28 | 云知声智能科技股份有限公司 | Single-channel voice dereverberation method and device |
| CN111540372A (en) * | 2020-04-28 | 2020-08-14 | 北京声智科技有限公司 | Method and device for multi-microphone array noise reduction processing |
| CN111540372B (en) * | 2020-04-28 | 2023-09-12 | 北京声智科技有限公司 | Method and device for noise reduction processing of multi-microphone array |
| CN111933170A (en) * | 2020-07-20 | 2020-11-13 | 歌尔科技有限公司 | Voice signal processing method, device, equipment and storage medium |
| CN111933170B (en) * | 2020-07-20 | 2024-03-29 | 歌尔科技有限公司 | Voice signal processing method, device, equipment and storage medium |
| CN115065422A (en) * | 2021-07-26 | 2022-09-16 | 中国计量科学研究院 | System and method for evaluating communication quality in reverberant room |
| CN114205731A (en) * | 2021-12-08 | 2022-03-18 | 随锐科技集团股份有限公司 | Speaker area detection method, device, electronic equipment and storage medium |
| CN114205731B (en) * | 2021-12-08 | 2023-12-26 | 随锐科技集团股份有限公司 | Speaker area detection method, speaker area detection device, electronic equipment and storage medium |
| CN117316175A (en) * | 2023-11-28 | 2023-12-29 | 山东放牛班动漫有限公司 | Intelligent encoding storage method and system for cartoon data |
| CN117316175B (en) * | 2023-11-28 | 2024-01-30 | 山东放牛班动漫有限公司 | Intelligent encoding storage method and system for cartoon data |
| CN117318671A (en) * | 2023-11-29 | 2023-12-29 | 有研(广东)新材料技术研究院 | Self-adaptive filtering method based on fast Fourier transform |
| CN117318671B (en) * | 2023-11-29 | 2024-04-23 | 有研(广东)新材料技术研究院 | Self-adaptive filtering method based on fast Fourier transform |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108172231B (en) | 2021-07-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108172231A (en) | A kind of dereverberation method and system based on Kalman filtering | |
| US7313518B2 (en) | Noise reduction method and device using two pass filtering | |
| Krueger et al. | Model-based feature enhancement for reverberant speech recognition | |
| CN109727604A (en) | Frequency domain echo cancel method and computer storage media for speech recognition front-ends | |
| Yoshioka et al. | Integrated speech enhancement method using noise suppression and dereverberation | |
| US20100198588A1 (en) | Signal bandwidth extending apparatus | |
| WO2009110574A1 (en) | Signal emphasis device, method thereof, program, and recording medium | |
| US8218780B2 (en) | Methods and systems for blind dereverberation | |
| JP2004502977A (en) | Subband exponential smoothing noise cancellation system | |
| CN111213359B (en) | Echo canceller and method for echo canceller | |
| RU2768514C2 (en) | Signal processor and method for providing processed noise-suppressed audio signal with suppressed reverberation | |
| US11483651B2 (en) | Processing audio signals | |
| US10937418B1 (en) | Echo cancellation by acoustic playback estimation | |
| Lei et al. | Deep neural network based regression approach for acoustic echo cancellation | |
| US7890319B2 (en) | Signal processing apparatus and method thereof | |
| JP6190373B2 (en) | Audio signal noise attenuation | |
| Lollmann et al. | A blind speech enhancement algorithm for the suppression of late reverberation and noise | |
| Compernolle | DSP techniques for speech enhancement | |
| Sehr et al. | Towards robust distant-talking automatic speech recognition in reverberant environments | |
| CN113160842B (en) | MCLP-based voice dereverberation method and system | |
| Kamarudin et al. | Acoustic echo cancellation using adaptive filtering algorithms for Quranic accents (Qiraat) identification | |
| CN114566179A (en) | Time delay controllable voice noise reduction method | |
| Razani et al. | A reduced complexity MFCC-based deep neural network approach for speech enhancement | |
| KR101537653B1 (en) | Method and system for noise reduction based on spectral and temporal correlations | |
| Dionelis | On single-channel speech enhancement and on non-linear modulation-domain Kalman filtering |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |