CN106782595A - A kind of robust blocking matrix method for reducing voice leakage - Google Patents
A kind of robust blocking matrix method for reducing voice leakage Download PDFInfo
- Publication number
- CN106782595A CN106782595A CN201611218157.7A CN201611218157A CN106782595A CN 106782595 A CN106782595 A CN 106782595A CN 201611218157 A CN201611218157 A CN 201611218157A CN 106782595 A CN106782595 A CN 106782595A
- Authority
- CN
- China
- Prior art keywords
- signal
- module
- blocking matrix
- voice signal
- voice
- 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
- 239000011159 matrix material Substances 0.000 title claims abstract description 94
- 230000000903 blocking effect Effects 0.000 title claims abstract description 87
- 238000000034 method Methods 0.000 title claims abstract description 56
- 238000009499 grossing Methods 0.000 claims description 9
- 230000008569 process Effects 0.000 claims description 5
- 238000001228 spectrum Methods 0.000 claims description 4
- 230000010354 integration Effects 0.000 claims description 3
- 230000008030 elimination Effects 0.000 abstract description 9
- 238000003379 elimination reaction Methods 0.000 abstract description 9
- 230000002708 enhancing effect Effects 0.000 abstract description 5
- 230000005236 sound signal Effects 0.000 abstract 1
- 238000005516 engineering process Methods 0.000 description 6
- 230000000694 effects Effects 0.000 description 5
- 230000006978 adaptation Effects 0.000 description 4
- 230000008859 change Effects 0.000 description 4
- 230000006870 function Effects 0.000 description 4
- 230000004048 modification Effects 0.000 description 4
- 238000012986 modification Methods 0.000 description 4
- 230000003044 adaptive effect Effects 0.000 description 3
- 238000004891 communication Methods 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000003993 interaction Effects 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 241000208340 Araliaceae Species 0.000 description 1
- 238000012935 Averaging Methods 0.000 description 1
- 235000005035 Panax pseudoginseng ssp. pseudoginseng Nutrition 0.000 description 1
- 235000003140 Panax quinquefolius Nutrition 0.000 description 1
- 230000002411 adverse Effects 0.000 description 1
- 230000004075 alteration Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 239000004568 cement Substances 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 235000008434 ginseng Nutrition 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000008439 repair process Effects 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000000926 separation method Methods 0.000 description 1
- 239000007787 solid Substances 0.000 description 1
- 230000003595 spectral effect Effects 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
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
- G10L21/0216—Noise filtering characterised by the method used for estimating noise
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
- G10L15/00—Speech recognition
- G10L15/20—Speech recognition techniques specially adapted for robustness in adverse environments, e.g. in noise, of stress induced speech
-
- 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
- G10L2021/02161—Number of inputs available containing the signal or the noise to be suppressed
- G10L2021/02166—Microphone arrays; Beamforming
Abstract
The invention discloses a kind of robust blocking matrix method for reducing voice leakage, including：Input audio signal；Targeted voice signal is obtained from voice signal using fixed beam module；Targeted voice signal is eliminated from voice signal using blocking matrix module, noise signal is obtained；Using the prior probability that there is targeted voice signal in fixed beam module estimated noise signal；Blocking matrix module updates noise signal according to prior probability, eliminates targeted voice signal, the noise signal after being updated present in noise signal；Using the noise signal that blocking matrix module is exported is eliminated in the targeted voice signal that cancellation module is exported from fixed beam module, form output signal and exported.The present invention is before the noise signal remained in eliminating targeted voice signal using cancellation module, the blocking matrix parameter of blocking matrix module is updated in advance, to eliminate the targeted voice signal omitted in noise signal, the function of the elimination targeted voice signal of enhancing blocking matrix module.
Description
Technical field
The present invention relates to field of speech recognition, more particularly to a kind of robust blocking matrix method for reducing voice leakage.
Background technology
Speech enhancement technique based on microphone array has been widely used for communication, man-machine interaction, speech recognition system
In, wherein generalized sidelobe elimination (GSC) method is most widely used, and this method is easily achieved and performance is fine.Wherein GSC is divided into
Upper and lower two paths, upper path is the reference signal that fixed beam module (FBF) is used to estimate target voice, and underpass is obstruction
Matrix module (BM) and cancellation module (MC), the wherein noise for eliminating the residual in fixed beam, blocking matrix module are used
Noise signal is obtained in elimination targeted voice signal.
From the point of view of many practice systems, it is exactly voice in BM modules is revealed to be easiest to allow the hydraulic performance decline of GSC, also
It is that BM does not block targeted voice signal, causes the voice signal for subtracting each other and balancing out leakage with the voice signal in FBF.Pass
The conventional self adaptation BM of BM designs of system directly uses difference matrix.Because the error of microphone array system, or target side
To estimation there is error, then difference matrix performance will be had a greatly reduced quality, and self adaptation BM will be walked by adaptive weight value updating
Influence long, the convergence of self adaptation is one than larger problem.
The content of the invention
The technical problems to be solved by the invention are to provide a kind of robust blocking matrix method for reducing voice leakage, can be big
The reduction voice leak case of amplitude.
To realize above-mentioned technique effect, the invention discloses a kind of robust blocking matrix method for reducing voice leakage, bag
Include step：
One voice signal is provided；
The voice signal is input into the fixed beam module and blocking matrix module of generalized side lobe structure, institute
Stating generalized side lobe structure has the first path and alternate path in parallel, and the fixed beam module is located at described first and leads to
Road, the blocking matrix module is located at the alternate path；The alternate path is additionally provided with cancellation module, the cancellation module
Input is connected with the output of the blocking matrix module, the output phase of the output of the cancellation module and the fixed beam module
Even；
Targeted voice signal is obtained from the voice signal of input using the fixed beam module, and exports described
Targeted voice signal；
Targeted voice signal is eliminated from the voice signal of input using the blocking matrix module, to obtain noise
Signal；
Using the prior probability that there is targeted voice signal in the fixed beam module estimation noise signal；
The blocking matrix module updates the noise signal according to the prior probability, is deposited in the elimination noise signal
Targeted voice signal, noise signal after being updated simultaneously exports the noise signal after updating；
Using eliminating the resistance in the targeted voice signal that the cancellation module is exported from the fixed beam module
The noise signal of plug matrix module output, forms output signal and is exported.
The present invention makes it have following beneficial effect as a result of above technical scheme：Cancellation module is being utilized to solid
The noise signal of the targeted voice signal and the output of blocking matrix module of determining beam steering module output is offseted, to eliminate target language
Before the noise signal remained in message number, the noise signal to the output of blocking matrix module carries out the presence of target language message in advance
Number probability priori, the blocking matrix parameter of blocking matrix module is updated, to eliminate the target language message omitted in noise signal
Number, the function of the elimination targeted voice signal of enhancing blocking matrix module, it is to avoid because blocking matrix module is not by target voice
Completely, the targeted voice signal in causing it with fixed beam module subtracts each other and balances out the target language message of leakage signal jam
Number, reach the situation that voice leakage is greatly reduced.
The robust blocking matrix method for reducing voice leakage is further improved and is, the voice of the voice signal
Bifurcation model is：
H0：X=N
H1：X=S+N (formula one)
Wherein, H0State representation only exists the state of noise, and N represents noise signal, H1State representation exist noise signal and
The state of targeted voice signal, S is targeted voice signal.
The robust blocking matrix method for reducing voice leakage is further improved and is, the voice signal is Mike
Wind input signal, the fixed beam module obtains targeted voice signal from the microphone input signal of input and gives defeated
Go out；The output Y of the fixed beam moduleFBFFor：
Wherein, M is microphone number, xiIt is i-th microphone input signal, w is the weight of fixed beam module, wiIt is
I-th weight of fixed beam.
It is described reduce voice leakage robust blocking matrix method further improve be, using postpone summation method or
Minimum secondary lobe class method is calculated the weight w of the fixed beam module.
It is described reduce voice leakage robust blocking matrix method further improve be, the blocking matrix module from
Targeted voice signal is eliminated in the microphone input signal of input, to obtain noise signal and be exported；The obstruction
The output Z of matrix module is：
Z=B*X (formula three)
Wherein, Z=[z1z2…zN], it is the output signal of blocking matrix module；X=[x1x2…xM], it is microphone input
Signal；B is the blocking matrix of blocking matrix module.
The robust blocking matrix method for reducing voice leakage is further improved and is, using the fixed beam mould
The output Y of blockFBFIn condition prior probably estimation go out in noise signal Z to exist the prior probability of targeted voice signal, including step
Suddenly：
Y is estimated with control recursive average algorithmFBFIt is middle there is targeted voice signal probability P (H1 | YFBF), to obtain Z
The middle prior probability P (H that there is targeted voice signal1)：
P(H1)k=λ P (H1)k-1+(1-λ)P(H1|YFBF) (formula four)
Wherein,
H1It is voice existence, λ is smoothing factor, and k is frame number；
Then in the absence of the prior probability P (H of targeted voice signal in Z0), tried to achieve by below equation
P(H0)=1-P (H1).(formula six)
The robust blocking matrix method for reducing voice leakage is further improved and is, the blocking matrix module root
The noise signal is updated according to the prior probability, targeted voice signal present in the noise signal is eliminated, is updated
The process of noise signal afterwards, including step：
Step one：Solve the condition prior probability P (H1 | Z) that there is targeted voice signal in Z
A, solution posteriori SNR γ
Wherein,It is the estimation of noise signal；
B, prior weight ε is solved using decision-directed method
Wherein, η is smoothing factor, value 0.92, γoldIt is the posteriori SNR of former frame, GH1It is H1Voice under state
Gain, MAX is to take big function；
There is likelihood score GLR in c, solution voice
Wherein,
D, solving condition prior probability P (H1 | BM)
Step 2：Amendment signal to noise ratio and renewal speech gain
A, using prior probability P (H1) amendment signal to noise ratio
Wherein,It is revised posteriori SNR,It is revised prior weight；
B, renewal speech gain GH1,
Wherein,
Exp is index operator, and e is natural constant, and x is integration variable；
Step 3：Estimate dynamic noise smoothing factor
Wherein, α values are 0.92；
Step 4：Estimate noise
Wherein, E is desired operation, is estimated using equation below：
Wherein, k is frame number, and ε represents prior weight, P (H0| BM)=1-P (H1 | BM)；
Step 5：Calculate speech gain
Speech gain Gain after updating is estimated using the log-magnitude spectrum method of estimation of optimal amendment
Wherein, Gmin is gain floor constraint when voice does not exist, and Gmin values are 0.01,It is in H1Shape
Speech gain when state,It is in H0Speech gain when state；
Step 6：It is calculated the noise signal Z ' after updating
Z '=Z* (1-Gain).(formula 17)
Brief description of the drawings
Fig. 1 is a kind of high-level schematic functional block diagram of the robust blocking matrix method for reducing voice leakage of the present invention.
Specific embodiment
Embodiments of the present invention are illustrated below by way of specific instantiation, those skilled in the art can be by this specification
Disclosed content understands other advantages of the invention and effect easily.The present invention can also be by specific realities different in addition
The mode of applying is embodied or practiced, the various details in this specification can also based on different viewpoints with application, without departing from
Various modifications or alterations are carried out under spirit of the invention.
The main task of speech enhancement technique is to suppress ambient noise and interference, so as to strengthen subsequent treatment be made an uproar for input
The robustness of sound.In traditional Single Channel Speech Enhancement Algorithm, due to there was only single analog signal input, without any with reference to letter
Number, noise, enhancing voice can only be suppressed in the statistical property of time domain and frequency domain using noisy speech signal.But voice signal exists
Often all it is submerged in time domain and frequency domain in the middle of noise and interference, it is difficult to accurate disconnected, therefore algorithm effect lifting
Space is relatively small.The speech enhancement technique that is introduced as of microphone array opens a brand-new thinking, and it utilizes target language
The difference of sound and interference on locus, and the correlation between each microphone signal, are calculated by Wave beam forming
Method ambient noise upper to arrival bearing and speech Separation and interference suppress, so as to strengthen voice, have been increasingly becoming voice
Strengthen the focus of area research.
In existing beamforming algorithm, using generalized side lobe (Generalized Sidelobe Canceller,
Abbreviation GSC) adaptive beam-forming algorithm of structure occupies an important position.
Below in conjunction with the accompanying drawings and specific embodiment the present invention is further detailed explanation.
Refer to shown in Fig. 1, Fig. 1 is that the functional module of the robust blocking matrix method that the present invention reduces voice leakage is illustrated
Figure, is also the schematic diagram of generalized side lobe structure.
Generalized side lobe structure (GSC) is divided into upper and lower two paths：First path 101 and alternate path 102, this first
Path 101 is parallel with one another with the alternate path 102, and in figure, the first path 101 is located at upper path, and alternate path 102 is located at upper
Path.Mainly there is a fixed beam module (fixed beam fomer, abbreviation FBF) 11, in generalized side lobe structure
Blocking matrix module (Blocking Matrix, abbreviation BM) 12 and a cancellation module (Multiplc-input Canceller letters
Claim MC) 13 3 functional modules.Wherein, fixed beam module (FBF) 11 is located at the first path 101, blocking matrix module (BM)
12 and cancellation module (MC) 13 be located at alternate path 102.The input of fixed beam module (FBF) 11 and blocking matrix module (BM)
12 input is connected, and the output of blocking matrix module (BM) 12 is connected with the input of cancellation module (MC) 13, cancellation module (MC)
13 output is connected with the output of fixed beam module (FBF) 11, and the output in cancellation module (MC) 13 and fixed beam mould
Carried out at the crossed node of the output of block (FBF) 11 " +/- " (and/or logical operation).
Wherein, fixed beam module (FBF) is used to estimate the reference signal of target voice, the filter that FBF is fixed using coefficient
Ripple device is filtered to original each channel signal, and filtered each road signal is added, so as to be different from mesh to arrival bearing
The interference and noise for marking voice signal are suppressed, and realize the first enhancing of targeted voice signal.
Blocking matrix module (BM) is used to eliminate targeted voice signal and obtain noise signal, and BM is exported as ginseng using FBF
Signal is examined, adaptive-filtering is carried out to each channel original signal, it is therefore an objective to target voice composition therein is removed, so as to obtain N
The noise signal (N is the number of microphone) on road, the sef-adapting filter of the process can use the CCAF (self adaptations that coefficient is defined
Wave filter).
Finally, cancellation module (MC) is used to eliminate the noise of the residual in fixed beam, and MC is using this N roads noise letter above
Number, further adaptive noise reduction treatment is done to FBF outputs, targeted voice signal is strengthened again, it is final so as to obtain
Output, the sef-adapting filter of the process can use NCAF (sef-adapting filter that scope is defined).
The present invention is eliminated in (GSC) method for current generalized sidelobe, because its blocking matrix (BM) module is not by mesh
Mark voice signal blocks completely, causes the targeted voice signal being secured in beam steering module (FBF) to subtract each other and balance out leakage
Targeted voice signal problem, the present invention provide it is a kind of reduce voice leakage robust blocking matrix method, to reach significantly
Degree ground reduces voice leakage problem, lifts the speech enhan-cement effect of generalized sidelobe removing method, meets more superior, higher standard
Communication, man-machine interaction, speech recognition system etc..
The concrete methods of realizing that the present invention reduces the robust blocking matrix method of voice leakage is as follows：
S001：A voice signal is provided, the voice signal is containing noisy voice signal；
S002：The voice signal is input into the fixed beam module 11 (FBF) and obstruction square of generalized side lobe structure
In array module 12 (BM), generalized side lobe structure has the first path 101 and alternate path 102 in parallel, fixed beam mould
Block 11 is located at the first path 101, and blocking matrix module 12 is located at alternate path 102；Alternate path 102 is additionally provided with cancellation module
(MC) 13, the input of cancellation module 13 is connected with the output of blocking matrix module 12, the output of cancellation module 13 and fixed beam
The output of module 11 is connected；
S003：Targeted voice signal is obtained from the voice signal of input using fixed beam module 11, and exports target
Voice signal；
S004：Targeted voice signal is eliminated from the voice signal of input using blocking matrix module 12, to obtain noise
Signal；
S004：Using the prior probability that there is targeted voice signal in the estimated noise signal of fixed beam module 11；
S005：Blocking matrix module 12 updates noise signal according to prior probability, eliminates target present in noise signal
Voice signal, noise signal after being updated simultaneously exports the noise signal after updating；
S006：Using elimination blocking matrix mould in the targeted voice signal that cancellation module 13 is exported from fixed beam module 11
The noise signal of the output of block 12, forms output signal and is exported.
As a example by below using a microphone input signal as voice signal, the microphone input signal is input into by broad sense
Valve is offseted in structure, and carries out voice increasing to the microphone input signal of the input using robust blocking matrix method of the invention
By force, it is specific as follows：
(1) it is input into microphone input signal；
The voice bifurcation model of the microphone input signal is：
H0：X=N
H1：X=S+N (formula one)
Wherein, H0State representation only exists the state of noise, and N represents noise signal, H1State representation exist noise signal and
The state of targeted voice signal, S is targeted voice signal.
(2) fixed beam module 11 (FBF) obtains targeted voice signal and gives from the microphone input signal of input
Output；
The output Y of fixed beam module (FBF)FBFFor：
Wherein, M is microphone number, xiIt is i-th microphone input signal, w is the weight of fixed beam module, wiIt is
I-th weight of fixed beam；The weight w of fixed beam module can be calculated using delay summation method or minimum secondary lobe class method
Obtain.
(3) blocking matrix module (BM) eliminates targeted voice signal from the microphone input signal of input, to be made an uproar
Acoustical signal is simultaneously exported；
The output Z of blocking matrix module (BM) is：
Z=B*X (formula three)
Wherein, Z=[z1z2…zN], it is the output signal (noise signal) of blocking matrix module；X=[x1x2…xM], it is
Microphone input signal；B is the blocking matrix of blocking matrix module, and blocking matrix is often tried to achieve with the method for difference.
(4) using the output Y of fixed beam module (FBF)FBFIn condition prior probably estimation go out blocking matrix module
(BM) there is the prior probability P (H of targeted voice signal in output signal Z (noise signal)1), it is specific as follows：
Y is estimated with control recursive average algorithmFBFIt is middle there is targeted voice signal probability P (H1 | YFBF), to obtain Z
The middle prior probability P (H that there is targeted voice signal1)：
P(H1)k=λ P (H1)k-1+(1-λ)P(H1|YFBF) (formula four)
Wherein,
H1It is voice existence, λ is smoothing factor, and k is frame number；
Control recursive average algorithm see " Israel Cohen Noise Spectrum Estimation in
Adverse Environments：Improved Minima Controlled Recursive Averaging”——IEEE
TRANSACTIONS ON SPEECH AND AUDIO PROCESSING, VOL.11, NO.5, SEPTEMBER 2003/
Page466-475.Describe the principle of operation of control recursive average algorithm in article in detail.
Now, in the absence of the priori of targeted voice signal in the output signal Z (noise signal) of blocking matrix module (BM)
Probability P (H0), tried to achieve by below equation
P(H0)=1-P (H1).(formula six)
(5) the prior probability P (H that blocking matrix module (BM) is estimated according to fixed beam module (FBF)1) update resistance
The noise signal of plug matrix module (BM) output, to eliminate the targeted voice signal still suffered from noise signal, after being updated
Noise signal, detailed process is as follows：
Step one：Solve the condition prior probability P (H1 | Z) that there is targeted voice signal in Z
A, solution posteriori SNR γ
Wherein,It is the estimation of noise signal；
B, prior weight ε is solved using decision-directed method
Wherein, η is smoothing factor, value 0.92, γoldIt is the posteriori SNR of former frame, GH1It is H1Voice under state
Gain, MAX is to take big function；
There is likelihood score GLR in c, solution voice
Wherein,
Exp is index transport symbol.
D, solving condition prior probability P (H1 | BM)
Step 2：Amendment signal to noise ratio and renewal speech gain
A, using prior probability P (H1) amendment signal to noise ratio
Wherein,It is revised posteriori SNR,It is revised prior weight；
B, renewal speech gain GH1,
Wherein,
Exp is index operator, and e is natural constant, and x is integration variable；
Step 3：Estimate dynamic noise smoothing factor
Wherein, α values are 0.92；
Step 4：Estimate noise
Wherein, E is desired operation, is estimated using equation below：
Wherein, k is frame number, and ε represents prior weight, P (H0| BM)=1-P (H1 | BM)；
Step 5：Calculate speech gain
Speech gain Gain after updating is estimated using log-magnitude Power estimation (OM-LSA) method of optimal amendment
Wherein, Gmin is gain floor constraint when voice does not exist, and Gmin values are 0.01 (- 20dB), -20dB=
(10*log10 (0.01)) dB, dB is the unit of decibel；It is in H1Speech gain when state,It is in H0Speech gain when state, but in order to prevent decay excessive, generally by GH0It is changed to Gmin conducts
H0When gain floor constraint
OM-LSA (the log-magnitude spectrums of the Optimally-Modified Log Spectral optimal amendments of Amplitude
Estimate) method see " Irael Cohen, Baruch BerdugoSpeech enhancement for non-
Stationary noise environment " --- the .February1990 of J.A couSsot.c Am 87 (2),
1990Acoustical Society of America/Page820-857.Describe the reality of OM-LSA methods in article in detail
Existing principle.
Step 6：It is calculated the noise signal Z ' after updating
Z '=Z* (1-Gain).(formula 17)
Using the above method, blocking matrix module updates noise signal according to prior probability, exists in elimination noise signal
Targeted voice signal, finally output update after noise signal.
(6) elimination blocking matrix module is defeated in the targeted voice signal that utilization cancellation module is exported from fixed beam module
The noise signal for going out, forms output signal and is exported.
The present invention reduces the robust blocking matrix method of voice leakage by utilizing cancellation module to fixed beam module
The noise signal of targeted voice signal and blocking matrix the module output of output is offseted, residual in targeted voice signal to eliminate
Before the noise signal stayed, the noise signal to the output of blocking matrix module carries out the probability elder generation that there is targeted voice signal in advance
Test, update the blocking matrix parameter of blocking matrix module, to eliminate the targeted voice signal omitted in noise signal, enhancing obstruction
The function of the elimination targeted voice signal of matrix module, it is to avoid because blocking matrix module has not blocked targeted voice signal
Entirely, the targeted voice signal in causing it with fixed beam module subtracts each other and balances out the targeted voice signal of leakage, reaches big
Amplitude reduces the situation of voice leakage.
It should be noted that structure, ratio, size depicted in this specification institute accompanying drawings etc., is only used to coordinate
Content disclosed in bright book, so that those skilled in the art understands and reads, is not limited to enforceable limit of the invention
Fixed condition, therefore do not have technical essential meaning, the modification of any structure, the change of proportionate relationship or the adjustment of size, not
Under influence effect that can be generated of the invention and the purpose to be reached, all should still fall obtained in disclosed technology contents
In the range of covering.Meanwhile, in this specification it is cited as " on ", D score, "left", "right", " centre " and " one " etc.
Term, is merely convenient to understanding for narration, and is not used to limit enforceable scope of the invention, the change of its relativeness or tune
It is whole, under without essence change technology contents, when being also considered as enforceable category of the invention.
The above is only presently preferred embodiments of the present invention, and any formal limitation is not done to the present invention, though
So the present invention is disclosed above with preferred embodiment, but is not limited to the present invention, any to be familiar with this professional technology people
Member, not departing from the range of technical solution of the present invention, when making a little change using the technology contents of the disclosure above or repair
The Equivalent embodiments for equivalent variations are adornd, as long as being the content without departing from technical solution of the present invention, according to technology reality of the invention
Any simple modification, equivalent variations and modification that confrontation above example is made, still fall within the scope of technical solution of the present invention
It is interior.
Claims (7)
1. a kind of robust blocking matrix method for reducing voice leakage, including step：
One voice signal is provided；
The voice signal is input into the fixed beam module and blocking matrix module of generalized side lobe structure, it is described wide
Adopted sidelobe cancellation structure has the first path and alternate path in parallel, and the fixed beam module is located at first path,
The blocking matrix module is located at the alternate path；The alternate path is additionally provided with cancellation module, the cancellation module it is defeated
Enter and be connected with the output of the blocking matrix module, the output phase of the output of the cancellation module and the fixed beam module
Even；
Targeted voice signal is obtained from the voice signal of input using the fixed beam module, and exports the target
Voice signal；
Targeted voice signal is eliminated from the voice signal of input using the blocking matrix module, to obtain noise letter
Number；
Using the prior probability that there is targeted voice signal in the fixed beam module estimation noise signal；
The blocking matrix module updates the noise signal according to the prior probability, eliminates present in the noise signal
Targeted voice signal, noise signal after being updated simultaneously exports the noise signal after updating；
Using eliminating the obstruction square in the targeted voice signal that the cancellation module is exported from the fixed beam module
The noise signal of array module output, forms output signal and is exported.
2. it is as claimed in claim 1 to reduce the robust blocking matrix method that voice is revealed, it is characterised in that the voice signal
Voice bifurcation model be：
H0：X=N
H1：X=S+N (formula one)
Wherein, H0State representation only exists the state of noise, and N represents noise signal, H1There is noise signal and target in state representation
The state of voice signal, S is targeted voice signal.
3. it is as claimed in claim 2 to reduce the robust blocking matrix method that voice is revealed, it is characterised in that the voice signal
It is microphone input signal, the fixed beam module obtains targeted voice signal and gives from the microphone input signal of input
To export；The output Y of the fixed beam moduleFBFFor：
Wherein, M is microphone number, xiIt is i-th microphone input signal, w is the weight of fixed beam module, wiIt is i-th
The weight of fixed beam.
It is 4. as claimed in claim 3 to reduce the robust blocking matrix method that voice is revealed, it is characterised in that：Sued for peace using delay
Method or minimum secondary lobe class method are calculated the weight w of the fixed beam module.
5. it is as claimed in claim 3 to reduce the robust blocking matrix method that voice is revealed, it is characterised in that the blocking matrix
Module eliminates targeted voice signal from the microphone input signal of input, to obtain noise signal and be exported；Institute
The output Z for stating blocking matrix module is：
Z=B*X (formula three)
Wherein, Z=[z1z2…zN], it is the output signal of blocking matrix module；X=[x1x2…xM], it is microphone input signal；
B is the blocking matrix of blocking matrix module.
6. it is as claimed in claim 5 to reduce the robust blocking matrix method that voice is revealed, it is characterised in that to utilize the fixation
The output Y of beam steering moduleFBFIn condition prior probably estimation go out in noise signal Z to exist the prior probability of targeted voice signal,
Including step：
Y is estimated with control recursive average algorithmFBFIt is middle there is targeted voice signal probability P (H1 | YFBF), deposit in Z with being obtained
In the prior probability P (H of targeted voice signal1)：
P(H1)k=λ P (H1)k-1+(1-λ)P(H1|YFBF) (formula four)
Wherein,
H1It is voice existence, λ is smoothing factor, and k is frame number；
Then in the absence of the prior probability P (H of targeted voice signal in Z0), tried to achieve by below equation
P(H0)=1-P (H1).(formula six)
7. it is as claimed in claim 6 to reduce the robust blocking matrix method that voice is revealed, it is characterised in that the blocking matrix
Module updates the noise signal according to the prior probability, eliminates targeted voice signal present in the noise signal, obtains
The process of the noise signal after to renewal, including step：
Step one：Solve the condition prior probability P (H1 | Z) that there is targeted voice signal in Z
A, solution posteriori SNR γ
Wherein,It is the estimation of noise signal；
B, prior weight ε is solved using decision-directed method
Wherein, η is smoothing factor, value 0.92, γoldIt is the posteriori SNR of former frame, GH1It is H1Voice under state increases
Benefit, MAX is to take big function；
There is likelihood score GLR in c, solution voice
Wherein,
D, solving condition prior probability P (H1 | BM)
Step 2：Amendment signal to noise ratio and renewal speech gain
A, using prior probability P (H1) amendment signal to noise ratio
Wherein,It is revised posteriori SNR,It is revised prior weight；
B, renewal speech gain GH1,
Wherein,
Exp is index operator, and e is natural constant, and x is integration variable；
Step 3：Estimate dynamic noise smoothing factor
Wherein, α values are 0.92；
Step 4：Estimate noise
Wherein, E is desired operation, is estimated using equation below：
Wherein, k is frame number, and ε represents prior weight, P (H0| BM)=1-P (H1|BM)；
Step 5：Calculate speech gain
Speech gain Gain after updating is estimated using the log-magnitude spectrum method of estimation of optimal amendment
Wherein, Gmin is gain floor constraint when voice does not exist, and Gmin values are 0.01,It is in H1During state
The speech gain of time,It is in H0Speech gain when state；
Step 6：It is calculated the noise signal Z ' after updating
Z '=Z* (1-Gain).(formula 17)
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611218157.7A CN106782595B (en) | 2016-12-26 | 2016-12-26 | Robust blocking matrix method for reducing voice leakage |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201611218157.7A CN106782595B (en) | 2016-12-26 | 2016-12-26 | Robust blocking matrix method for reducing voice leakage |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106782595A true CN106782595A (en) | 2017-05-31 |
| CN106782595B CN106782595B (en) | 2020-06-09 |
Family
ID=58925084
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201611218157.7A Active CN106782595B (en) | 2016-12-26 | 2016-12-26 | Robust blocking matrix method for reducing voice leakage |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106782595B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107301869A (en) * | 2017-08-17 | 2017-10-27 | 珠海全志科技股份有限公司 | Microphone array sound pick-up method, processor and its storage medium |
| CN109473118A (en) * | 2018-12-24 | 2019-03-15 | 苏州思必驰信息科技有限公司 | Double-channel pronunciation Enhancement Method and device |
| CN111341340A (en) * | 2020-02-28 | 2020-06-26 | 重庆邮电大学 | Robust GSC method based on coherence and energy ratio |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR20080075362A (en) * | 2007-02-12 | 2008-08-18 | 인하대학교 산학협력단 | A method for obtaining an estimated speech signal in noisy environments |
| CN101976565A (en) * | 2010-07-09 | 2011-02-16 | 瑞声声学科技(深圳)有限公司 | Dual-microphone-based speech enhancement device and method |
| CN103106390A (en) * | 2011-11-11 | 2013-05-15 | 索尼公司 | Information processing apparatus, information processing method, and program |
| CN103236260A (en) * | 2013-03-29 | 2013-08-07 | 京东方科技集团股份有限公司 | Voice recognition system |
| KR20160116440A (en) * | 2015-03-30 | 2016-10-10 | 한국전자통신연구원 | SNR Extimation Apparatus and Method of Voice Recognition System |
-
2016
- 2016-12-26 CN CN201611218157.7A patent/CN106782595B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| KR20080075362A (en) * | 2007-02-12 | 2008-08-18 | 인하대학교 산학협력단 | A method for obtaining an estimated speech signal in noisy environments |
| CN101976565A (en) * | 2010-07-09 | 2011-02-16 | 瑞声声学科技(深圳)有限公司 | Dual-microphone-based speech enhancement device and method |
| CN103106390A (en) * | 2011-11-11 | 2013-05-15 | 索尼公司 | Information processing apparatus, information processing method, and program |
| CN103236260A (en) * | 2013-03-29 | 2013-08-07 | 京东方科技集团股份有限公司 | Voice recognition system |
| KR20160116440A (en) * | 2015-03-30 | 2016-10-10 | 한국전자통신연구원 | SNR Extimation Apparatus and Method of Voice Recognition System |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107301869A (en) * | 2017-08-17 | 2017-10-27 | 珠海全志科技股份有限公司 | Microphone array sound pick-up method, processor and its storage medium |
| CN109473118A (en) * | 2018-12-24 | 2019-03-15 | 苏州思必驰信息科技有限公司 | Double-channel pronunciation Enhancement Method and device |
| CN109473118B (en) * | 2018-12-24 | 2021-07-20 | 思必驰科技股份有限公司 | Dual-channel speech enhancement method and device |
| CN111341340A (en) * | 2020-02-28 | 2020-06-26 | 重庆邮电大学 | Robust GSC method based on coherence and energy ratio |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106782595B (en) | 2020-06-09 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108831495B (en) | Speech enhancement method applied to speech recognition in noise environment | |
| CN102938254B (en) | Voice signal enhancement system and method | |
| JP5444472B2 (en) | Sound source separation apparatus, sound source separation method, and program | |
| CN105280193B (en) | Priori signal-to-noise ratio estimation method based on MMSE error criterion | |
| CN110148420A (en) | A kind of audio recognition method suitable under noise circumstance | |
| CN101901602B (en) | Method for reducing noise by using hearing threshold of impaired hearing | |
| CN106251857A (en) | Sounnd source direction judgment means, method and mike directivity regulation system, method | |
| Knecht et al. | Neural network filters for speech enhancement | |
| CN106782595A (en) | A kind of robust blocking matrix method for reducing voice leakage | |
| Heymann et al. | Joint optimization of neural network-based WPE dereverberation and acoustic model for robust online ASR | |
| Li et al. | An investigation of spectral restoration algorithms for deep neural networks based noise robust speech recognition. | |
| CN107018470A (en) | A kind of voice recording method and system based on annular microphone array | |
| CN108200522A (en) | A kind of change regularization ratio normalization sub-band adaptive filtering method | |
| KR20190091061A (en) | An MVDR beamformer using a steering vector estimator based on an online complex Gaussian mixture model using recursive least squares | |
| CN105070295A (en) | Adaptive method, applied to echo cancellation, of active factor proportional sub band | |
| CN112530451A (en) | Speech enhancement method based on denoising autoencoder | |
| CN110191245B (en) | Self-adaptive echo cancellation method based on time-varying parameters | |
| CN112331226B (en) | Voice enhancement system and method for active noise reduction system | |
| Kalamani et al. | Modified noise reduction algorithm for speech enhancement | |
| CN105719658B (en) | Wavelet packet voice de-noising method based on new threshold function table and adaptive threshold | |
| CN107369456A (en) | Noise cancellation method based on generalized sidelobe canceller in digital deaf-aid | |
| EP3614696B1 (en) | Beam former, beam forming method and hearing aid system | |
| Priyanka et al. | Adaptive Beamforming Using Zelinski-TSNR Multichannel Postfilter for Speech Enhancement | |
| CN108462481A (en) | Ratio LMP filtering methods based on parameter adjustment under a kind of μ rule function | |
| CN110034747A (en) | The plural scaled symbol sef-adapting filter of robust |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| TA01 | Transfer of patent application right | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20170929 Address after: 200233 Shanghai City, Xuhui District Guangxi 65 No. 1 Jinglu room 702 unit 03 Applicant after: YUNZHISHENG (SHANGHAI) INTELLIGENT TECHNOLOGY CO.,LTD. Address before: 200233 Shanghai, Qinzhou, North Road, No. 82, building 2, layer 1198, Applicant before: SHANGHAI YUZHIYI INFORMATION TECHNOLOGY Co.,Ltd. |
|
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| PE01 | Entry into force of the registration of the contract for pledge of patent right |
Denomination of invention: A robust blocking matrix method for reducing speech leakage Effective date of registration: 20201201 Granted publication date: 20200609 Pledgee: Bank of Hangzhou Limited by Share Ltd. Shanghai branch Pledgor: YUNZHISHENG (SHANGHAI) INTELLIGENT TECHNOLOGY Co.,Ltd. Registration number: Y2020310000047 |
|
| PE01 | Entry into force of the registration of the contract for pledge of patent right | ||
| PC01 | Cancellation of the registration of the contract for pledge of patent right | ||
| PC01 | Cancellation of the registration of the contract for pledge of patent right |
Date of cancellation: 20220307 Granted publication date: 20200609 Pledgee: Bank of Hangzhou Limited by Share Ltd. Shanghai branch Pledgor: YUNZHISHENG (SHANGHAI) INTELLIGENT TECHNOLOGY CO.,LTD. Registration number: Y2020310000047 |
|
| PE01 | Entry into force of the registration of the contract for pledge of patent right | ||
| PE01 | Entry into force of the registration of the contract for pledge of patent right |
Denomination of invention: A Robust Blocking Matrix Method for Reducing Speech Leakage Effective date of registration: 20230210 Granted publication date: 20200609 Pledgee: Bank of Hangzhou Limited by Share Ltd. Shanghai branch Pledgor: YUNZHISHENG (SHANGHAI) INTELLIGENT TECHNOLOGY CO.,LTD. Registration number: Y2023310000028 |
|
| PC01 | Cancellation of the registration of the contract for pledge of patent right | ||
| PC01 | Cancellation of the registration of the contract for pledge of patent right |
Granted publication date: 20200609 Pledgee: Bank of Hangzhou Limited by Share Ltd. Shanghai branch Pledgor: YUNZHISHENG (SHANGHAI) INTELLIGENT TECHNOLOGY CO.,LTD. Registration number: Y2023310000028 |
|
| PE01 | Entry into force of the registration of the contract for pledge of patent right | ||
| PE01 | Entry into force of the registration of the contract for pledge of patent right |
Denomination of invention: A Robust Blocking Matrix Method for Reducing Speech Leakage Granted publication date: 20200609 Pledgee: Bank of Hangzhou Limited by Share Ltd. Shanghai branch Pledgor: YUNZHISHENG (SHANGHAI) INTELLIGENT TECHNOLOGY CO.,LTD. Registration number: Y2024310000165 |