CN102800322A - Method for estimating noise power spectrum and voice activity - Google Patents
Method for estimating noise power spectrum and voice activity Download PDFInfo
- Publication number
- CN102800322A CN102800322A CN2011101411375A CN201110141137A CN102800322A CN 102800322 A CN102800322 A CN 102800322A CN 2011101411375 A CN2011101411375 A CN 2011101411375A CN 201110141137 A CN201110141137 A CN 201110141137A CN 102800322 A CN102800322 A CN 102800322A
- Authority
- CN
- China
- Prior art keywords
- lambda
- overbar
- alpha
- voice
- probability
- 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
- 238000000034 method Methods 0.000 title claims abstract description 72
- 238000001228 spectrum Methods 0.000 title claims abstract description 72
- 230000000694 effects Effects 0.000 title claims abstract description 35
- 238000009826 distribution Methods 0.000 claims abstract description 14
- 238000007476 Maximum Likelihood Methods 0.000 claims abstract description 3
- 230000007704 transition Effects 0.000 claims description 33
- 238000001514 detection method Methods 0.000 claims description 22
- 238000009499 grossing Methods 0.000 claims description 13
- 239000000284 extract Substances 0.000 claims description 12
- 238000013459 approach Methods 0.000 claims description 10
- 230000006870 function Effects 0.000 claims description 8
- 238000012804 iterative process Methods 0.000 claims description 6
- 238000013461 design Methods 0.000 claims description 5
- 238000006467 substitution reaction Methods 0.000 claims description 4
- 230000005236 sound signal Effects 0.000 claims description 3
- 230000009466 transformation Effects 0.000 claims description 3
- 230000008569 process Effects 0.000 description 5
- 230000003044 adaptive effect Effects 0.000 description 3
- 230000006872 improvement Effects 0.000 description 3
- 230000011218 segmentation Effects 0.000 description 3
- 238000005516 engineering process Methods 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 206010038743 Restlessness Diseases 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 230000008878 coupling Effects 0.000 description 1
- 238000010168 coupling process Methods 0.000 description 1
- 238000005859 coupling reaction Methods 0.000 description 1
- 230000002950 deficient Effects 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 239000004615 ingredient Substances 0.000 description 1
- 230000009191 jumping Effects 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 230000000630 rising effect Effects 0.000 description 1
- 230000006641 stabilisation Effects 0.000 description 1
- 238000011105 stabilization Methods 0.000 description 1
- 238000013179 statistical model Methods 0.000 description 1
- 230000002123 temporal effect Effects 0.000 description 1
- 230000036962 time dependent Effects 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
Images
Landscapes
- Complex Calculations (AREA)
Abstract
The invention relates to a method for estimating the noise power spectrum and the voice activity. According to the method, the appear probability of a voice on a frequency sub band and power spectrum information of noise can be finally deduced according to the time sequence relevance of a sequential hidden markov model (SHMM) description language based on first-order regression on each frequency component. The method comprises the following steps of: 1) extracting a logarithmic amplitude spectrum envelop for a voice signal on each frequency component, and constructing a corresponding binary hidden markov model, wherein each state is represented by Gaussian distribution; 2) for a field of voice data, setting M frames of caches, storing the previous M frames of input signals into the caches, extracting M frames of logarithmic amplitude spectrums from the caches, and constructing an initialization model by adopting a maximum likelihood estimation algorithm; and 3) after the initialization model lambdaM is obtained, starting from the (M+1)th frame, gradually updating the HMM of each frequency band by adopting an incremental learning method, and sequentially performing recurrence to obtain a noise value and the appear probability of a voice signal.
Description
Technical field
The present invention relates to voice signal Treatment Technology field, specifically, the present invention relates to a kind of noise spectrum and estimate and the voice activity detection method based on sequential hidden Markov model.Wherein, it is on time dimension, to judge the algorithm whether voice occur that voice activity detects, and it can answer existence with the form of " being " or " denying ", also can describe the existence of voice with the voice probability of occurrence.
Background technology
Voice activity detects and noise power spectrum estimates it is the requisite ingredient of noise reduction algorithm; Their performance directly influences the performance of noise reduction algorithm; Particularly under abominable noise circumstance, their remote effect the performance of speech processing system (like speech recognition, words person's identification and speech recognizer).
Most voice application system is had in the face of ambient noise interference.Forefathers have proposed a lot of methods and have removed the interference of noise to voice system, and nearly all method all depends on the voice activity detection and noise power spectrum is estimated.These two modules exist contact closely, and their accuracy directly influences the whole noiseproof feature of system.Although traditional method of estimation is functional, but still have two places to be worth improving:
1, makes full use of the sequential correlativity of continuous speech/non-speech audio on a certain frequency component; Existing algorithm is abundant inadequately for the utilization of temporal correlation; They often adopt fairly simple single order recurrence smoother that the amplitude spectrum envelope is carried out smoothly, and the smoothing factor of smoother is fixed.And voice signal itself is exactly a segmentation signal stably, and its statistical nature comprises the sequential correlativity, and all along with the time is constantly changing, a fixing model can't reflect this time-varying characteristics.If we can adopt adaptive model that the sequential correlativity is carried out modeling, the performance of algorithm will get a promotion undoubtedly so.This method is not mentioned in documents and materials in the past.
2, the parameter adaptive of traditional sequential HMM adopts the recurrence average mode of high-order, and current HMM parameter set depends on the model in a moment, the current observed value and the observed value in a plurality of moment in past, and the mode calculated amount of this parametric regression is huge.If can return at following this high-order of the little situation of loss of significance and be reduced to the single order recurrence, so, the counting yield of algorithm will greatly improve.Based on the sequential HMM algorithm that single order returns, in documents and materials in the past, do not mention yet.
In addition, traditional solution is based on the mode of semi-supervised learning.At initial period, one system need make the hypothesis of " noise is initial ", supposes that promptly always there is one section non-speech audio in the beginning of sentence.This section non-speech audio is appreciated that the ground unrest sample into the manual work mark, from these mark samples, sets up the initial model of noise, and this is a kind of supervised learning method.Its defective is: this hypothesis is difficult to be met in some applications, such as starting with voice signal when sentence, will cause the initialization failure of noise model so, and it is all inaccurate to make speech detection and noise power spectrum estimate then.This initialized method is open in the patent of Chinese application number 201010178166.4.
Summary of the invention
The purpose of the object of the invention is; For being provided, a kind of noise spectrum based on sequential hidden Markov model estimates and the voice activity detection method; This method utilizes hidden Markov model that the sequential correlativity that voice signal exists on certain frequency component is carried out modeling, and the log power spectrum envelope on certain frequency component can be regarded a Markov chain as, the redirect between voice " appearance " and " not occurring " two states of this chain; For each state; Adopt a Gaussian distribution to describe the distribution of its power spectrum, again according to the forward direction factor of HMM, can derive voice certain the time frequency probability of occurrence.
For realizing the foregoing invention purpose; The invention provides a kind of noise power spectrum and estimate and the voice activity detection method that this method is described the sequential correlativity of voice on each frequency component based on the sequential hidden Markov model SHMM that single order returns, and adopt the mode of incremental learning to come progressively to upgrade SHMM; Finally; Deduce out the power spectrum information of the probability of occurrence on this frequency subband and the noise of voice, with the sequential statistical nature of accurate reflection voice, this method comprises the following steps:
1) on each frequency component, extracts logarithm amplitude spectrum envelope for voice signal; And set up the binary hidden Markov model of a correspondence, wherein, one-component is represented the distribution of speech energy; Another component is the distribution of noise energy, and each state is represented by Gaussian distribution;
2) for one section speech data, set the M frame buffer, deposit preceding M frame input signal in the buffer memory in, extract the logarithm amplitude spectrum of M frame in the buffer memory, adopt the maximum likelihood algorithm for estimating to set up an initialized model;
3) obtaining initialized model λ
MAfterwards, since the M+1 frame, adopt the method for incremental learning, by the HMM model of each frequency band of frame update, recursion obtains the probability of occurrence of noise figure and voice signal successively.
The concrete steps of this method comprise:
1) on each frequency component, extracts logarithm amplitude spectrum envelope for voice signal, for the logarithm amplitude spectrum time series x on the frequency component
l={ x
1, x
2..., x
l, set up a hidden Markov model s
l={ s
1, s
2..., s
l, s
t{ 0,1} is its corresponding status switch to ∈, and 1 expression voice go out present condition, and 0 expression noise goes out present condition, λ
lExpression is from sequence x
lIn the model parameter valuation of obtaining, so, for a given parameter set λ
l, corresponding observed value sequence x
lProbability density function can be expressed as:
Wherein, p (s
l| λ
l) expression status switch s
lThe prior probability that occurs, gaussian component is expressed as:
Here
The expression state transition probability,
Expression original state probability, p (x
l| λ
l, s
l) expression given state s
lWith parameter set λ
lSituation under observed value sequence x
lLikelihood score:
Wherein,
μ in this model
0, lBe exactly that we want the noise estimated, simultaneously, we can derive that voice signal occurs on certain frequency of l frame probability does
2) for one section speech data, set the M frame buffer, deposit preceding M frame input signal in the buffer memory in, extract the logarithm amplitude spectrum of M frame in the buffer memory, the HMM model of substitution step 1) is to hidden Markov model λ of initialization on each frequency
M, subscript M representes initialized time window length, l>=M;
3) obtaining initialized model λ
MAfterwards, since the M+1 frame, the HMM model adopts the method for incremental learning, and by frame update SHMM model, recursion obtains λ successively
lAnd draw noise figure μ
0, lWith the probability of occurrence of voice signal on certain frequency of l frame.
As a kind of improvement of technique scheme, the step of extracting frame amplitude spectrum in the described step 1) comprises:
At first, the digitized sound signal of this frame is done pre-service, establishing every frame length is the F point, and first zero padding is to N point, N>=F, N=2
j, j is an integer, and j>=8, carries out leaf transformation in the N point discrete Fourier, obtains discrete spectrum
Wherein, y
L, nN sampled point of l frame in the expression buffer memory, Y
L, kK Fourier transform value of i frame in the expression buffer memory (k=0,1 ..., N-1); So, its range value may be calculated
In the formula, b (r) is a windowed function.Described pre-service comprises windowing or/and pre-emphasis; Described windowed function adopts Hanning window or breathes out peaceful window.
As a kind of improvement of technique scheme, described step 2) in the initialization of HMM, concrete initialized step comprises on certain frequency:
Step 201): the method through cluster is divided into two types with M sample:
With
Wherein, M
0+ M
1=M, average bigger a type is with subscript (1) expression, another kind ofly representes with subscript (0); The method of the cluster described step 201) adopts non-supervision cluster of LBG or fuzzy clustering method;
Two types variance is respectively:
Two types initializes weights coefficient is:
The likelihood score of novel model of calculating,
And beginning interative computation; In following iterative process, old model parameter set is expressed as λ '
M, new model parameter is:
Before the beginning iteration,
L ' is set to a very big negative, the initialization forward direction factor,
After the initialization to the factor,
Step 202): calculate the forward direction factor:
Step 203): calculate back to the factor:
Step 204): calculating noise and voice probability of occurrence:
Step 206): calculate transition probability:
Step 207): calculate new initialization probability
Step 208): calculate new average
Step 209): the average to new retrains: μ '
1, M=max{ μ '
0, M, μ '
0, M+ δ }, wherein, δ is a constant, span is between 0 to 100;
Step 210): calculate new variance
Step 211): new variance is retrained κ '
1, M=max{ κ '
0, M, κ '
1, M;
Step 212): calculate new transition probability,
Step 213): the likelihood score of novel model of calculating
Step 214): if satisfy condition
termination of iterations; Wherein, ε is a very little numeral, if
iteration jumps to step 202).
In above-mentioned HMM modeling of parameters process, average, weight, variance and the transition probability to HMM retrains respectively.It is to be noted: in initialization procedure; Weight coefficient in effect that the transition probability is here brought into play and the patent 201010178166.4 is suitable; Because the weight coefficient in 201010178166.4 is used as denominator term in initialization procedure, so it must retrain in initialization procedure.And there is not this problem in the transition probability in this patent.
As a kind of improvement of technique scheme, the sequential renewal of the HMM in the described step 3) is to set up initialized model λ
MAfterwards, since the M+1 frame, adopt the method for incremental learning, by frame update HMM model, its iterative process can be expressed as: on each frequency, and known λ
lWith current observed value x
l, infer λ
L+1Carry out Fourier transform for the l+1 frame, obtain Y
L+1, k, wherein, 0≤k<N; On each frequency, calculate range value
For each frequency, following in the parameter update step of l+1 frame:
Step 301): calculate the forward direction factor,
Z ∈ 0,1};
Step 302): computing voice and noise probability of occurrence,
Step 303): the design conditions transition probability,
Step 304): calculating average noise voice probability of occurrence,
Step 305): rely on smoothing factor computing time,
Step 306): the computing mode average,
Step 307): the state average to new retrains: μ
1, l+1=max{ μ
1, l+1, μ
0, l+1+ δ }, l>=M;
Step 308): calculate new state variance,
Step 309): the new state variance is retrained κ
1, l+1=max{ κ
0, l+1, κ
1, l+1, l>=M;
Step 310): calculate average transition probability,
Step 311): the computing mode probability,
Step 312): the transition probability to new retrains, a
01, l=max{a
01, l, η }, a
00, l=1-a
01, l, a
10, l=max{a
10, l, η }, a
11, l=1-a
10, l, l>=M;
From above substep, obtained λ
L+1In all parameters, thereby obtained relevant voice probability of occurrence γ
L+1| λ l(1) and the power spectrum valuation μ of noise signal
0, l+1
The incremental learning method that HMM model in the described step 3) adopts comprises: recursion weight coefficient, recursion average and pass the vertebra variance;
Wherein, Described recursion average: in
formula;
is a smoothing factor that depends on the voice probability of occurrence, less than 1 but approach 1;
Described recursion variance:
Described recursion transition probability:
Perhaps a
Yz, l+1=β a
Yz, l+ (1-β) ξ
L+1| λ l(y, z); In the formula, β be one less than 1 but approach 1 smoothing factor, β=0.99 for example.
The parameter recurrence method of the described sequential hidden Markov model that returns based on single order is:
Calculate the forward direction factor of HMM:
Computing voice and noise probability of occurrence,
The design conditions transition probability,
Calculating average noise voice probability of occurrence,
Rely on smoothing factor computing time,
Computation of mean values,
Calculate new variance,
Calculate average transition probability,
Calculate transition probability,
In the technique scheme, the tied mechanism of the guarantee bigram statistics model sound and stable operation of employing comprises:
1) in the starting stage; When the average probability of occurrence of voice during less than certain fixed threshold ζ,
algorithm for estimating stops to fall generation.This constrains in the step 205 and implements.
2), the transfering state of model is retrained for to prevent that the state redirect of hidden Markov model from stopping.a
01,l=max{a
01,l,η},a
00,l=1-a
01,l,a
10,l=max{a
10,l,η},a
11,l=1-a
10,l,l≥M。This constrains in the step 312 and implements.
3) in tracing process, to the constraint of average, μ
1, l+1=max{ μ
1, l+1, μ
0, l+1+ δ }, l>=M.This constrains in the step 307 and implements.
4) to the constraint of variance, κ
1, l+1=max{ κ
0, l+1, κ
1, l+1, l>=M.This constrains in the step 309 and implements.
The present invention relates to a kind of based on sequential hidden Markov model (Sequential Hidden Markov Model; SHMM) noise power spectrum is estimated and the voice activity detection method, is comprised the following steps: 1) for the logarithm amplitude characteristic of voice signal on each frequency component, set up a SHMM model; 2) for one section speech data; Set the M frame buffer, deposit preceding M frame input signal in the buffer memory in, extract the logarithm amplitude spectrum of M frame in the buffer memory; The SHMM model of substitution step 1) carries out initialization, obtains initialized model λ
M3) obtaining initialized model λ
MAfterwards, since the M+1 frame, adopt the method for incremental learning, by frame update SHMM model.Noise states mean value in the model is exactly current noise estimation value, and the voice probability of occurrence in the estimation procedure is represented the activity of voice at time-frequency domain.The method of recursion is: according to current observed value x
lModel parameter collection λ with a last moment
L-1, the model parameter collection λ of estimation current time
lThus, obtain each the constantly noise power spectrum on certain frequency component and probability of voice appearance successively.The present invention is that spectrum is estimated and the tight coupling solution of voice activity detection, can strengthen the adaptability of voice application system to noise circumstance; The present invention does not rely on " noise is initial " and supposes; And the present invention can also provide the description of voice activity on the time-frequency two-dimensional space.This patent is development and come on the patent basis of the patent No. of having applied for 201010178166.4, owing to adopted the accurate modeling method of model more, the performance of this patent to be superior to 201010178166.4, but computation complexity is higher than 201010178166.4.
Compared with prior art, the present invention has following technique effect:
On certain frequency component, there is the sequential correlativity based on voice signal; The present invention utilizes hidden Markov model that this sequential correlativity is carried out modeling; Log power spectrum envelope on certain frequency component can be regarded a Markov chain as; The redirect between voice " appearance " and " not occurring " two states of this chain for each state, adopts a Gaussian distribution to describe the distribution of its power spectrum.In order to simplify calculating, the invention allows for the sequential HMM method for tracing that single order returns, its parameter is along with input signal constantly changes.Wherein the estimated value that the average of state is exactly a noise power spectrum " does not appear " in the voice of HMM, according to the forward direction factor of HMM, can derive voice certain the time frequency probability of occurrence.
The present invention is that a kind of voice activity detects and noise power spectrum is estimated tightly coupled scheme, can strengthen the adaptability of voice application system to noise circumstance; And the present invention can also provide the description of voice activity on the time-frequency two-dimensional space, helps noise is carried out further process of refinement.
Description of drawings
Fig. 1 noise spectrum of the present invention estimate with the voice activity detection method process flow diagram;
SHMM Noise Estimation algorithm of the present invention, classical minimum statistics algorithm (MS), minimum control return average algorithm (MCRA) and its raising version IMCRA effect comparison figure to Fig. 2 for instance has compared.
Embodiment
The present invention proposes a kind of noise power spectrum based on sequential hidden Markov model estimates and the voice activity detection method.
As shown in Figure 1, comprise the following steps:
1) for the logarithm amplitude characteristic of voice signal on each frequency, set up a HMM model, mathematic(al) representation is following:
expression state transition probability here;
expression original state probability, wherein gaussian component is expressed as:
Wherein, x
lRepresent the logarithm amplitude spectrum on certain frequency of l frame, z=0 representes that voice do not go out present condition, and z=1 expresses present condition.μ
Z, kAnd κ
Z, kRepresent average and variance respectively, parameter set λ
l={ μ
0, l, μ
1, l, κ
1, l, κ
0, l, a
01, l, a
10, l, a
00, l, a
11, l, π
0, π
1.
2) for one section speech data, set the M frame buffer, deposit preceding M frame input signal in the buffer memory in, extract the logarithm amplitude spectrum of M frame in the buffer memory, the GMM model of substitution step 1) carries out initialization, obtains initialized model λ
0, kInitialization procedure adopts constraint EM algorithm; M representes the length of initialization window.
3) obtaining initialized model λ
MAfterwards, since the M+1 frame, adopt the method for incremental learning, by frame update HMM model, recursion obtains λ successively
lAnd draw noise figure μ
0, lWith the probability of occurrence of voice signal on certain frequency of l frame.
I=1 wherein, 2,3 ...
Wherein, the incremental learning method of said GMM comprises recursion weight coefficient, recursion average and recursion variance;
Wherein forward direction factor recurrence method is:
Z ∈ 0,1}.
Voice and noise probability of occurrence recurrence method are:
Conditional transfer probability recurrence method is:
Average noise voice probability of occurrence recurrence method is:
Time-dependent smoothing factor recurrence method is:
State average recurrence method is:
The state variance recurrence method is:
Mean transferred probability recurrence method is:
The state probability recurrence method is:
The maximum characteristics of sequential hidden Markov model be can online tracking frequency component on the sequential correlativity that occurs of voice, it regards the general envelope of the power on certain frequency component as a Markov chain of between voice and non-voice state, switching.It adopts the mode of non-supervision to make up initial model.Particularly, it has following characteristics:
● owing to adopt HMM, can adopt the mode of Viterbi decoding, provide on the time series optimal estimation whether voice occur.
● at initial phase, do not rely on the initial hypothesis of noise, so the range of application that should invent is used more wide in range than one solution.
● voice activity is the two-dimensional signal of " time---frequency ", and other voice activity detection algorithm has only been described the existence of voice on time dimension.
In one embodiment, the carrier of unsupervised learning framework is binary hidden Markov model (Hidden MarkovModel is abbreviated as HMM).The distribution of one of them representation in components speech energy, another component are the distributions of noise energy.On each frequency component, extract logarithm amplitude spectrum envelope, and set up the HMM of a correspondence.At first adopt EM algorithm initialization HMM, adopt the mode of incremental learning to come progressively to upgrade HMM then.According to the HMM model, deduce out the power spectrum information of the probability of occurrence on this subband and the noise of voice respectively.In HMM modeling of parameters process, average, weight, variance and the transition probability to HMM retrains respectively.Wherein, for the sequential estimation method of HMM parameter, specifically comprise the calculating of recursion weight coefficient, recursion average and recursion variance and recursion.
The smoothing factor of sound probability of occurrence is less than 1 but approach 1.
2) recursion variance,
3) recursion transition probability;
perhaps
wherein β be one less than 1 but approach 1 smoothing factor, β=0.99 for example.
Below in conjunction with a preferred embodiment the present invention is done description further.
Principle of the present invention is following:
Noise estimation procedure parallel running on each frequency component, so, dispense frequency component index k in the following description.For the logarithm amplitude spectrum time series x of voice signal on each frequency component
l={ x
1, x
2..., x
l, set up a hidden Markov model, s
l={ s
1, s
2..., s
l, s
t{ 0,1} is its corresponding status switch to ∈, and 1 expression voice go out present condition, and 0 expression noise goes out present condition, λ
lExpression is from sequence x
lIn the model parameter valuation of obtaining, so for a given parameter set λ
l, corresponding observed value sequence x
lProbability density function can be expressed as:
Wherein, p (s
l| λ
l) expression status switch s
lThe prior probability gaussian component that occurs is expressed as:
Here
The expression state transitions
Expression original state probability, p (x
l| λ
l, s
l) expression given state s
lWith parameter set λ
lSituation under observed value sequence x
lLikelihood score:
Wherein
Here κ
Z, lExpression Gaussian distribution variance, μ
Z, lThe expression average, s
l=z, λ
l={ μ
0, l, μ
1, l, κ
1, l, κ
0, l, a
01, l, a
10, l, a
00, l, a
11, l, π
0, π
1, the initial probability π in the parameter set
zNot along with the time changes.
μ in this model
0, lBe exactly that we want the noise estimated.Simultaneously, our probability that can derive that voice signal occurs on certain frequency of l frame is γ
T| λ l(z)=p (s
t=z|x
t, λ
l).
Based on above-mentioned principle, according to one embodiment of present invention, said noise power spectrum is estimated and the voice activity detection method comprises the following steps:
Step 100: set the M frame buffer, deposit preceding M frame input signal in the buffer memory in, extract the amplitude spectrum of M frame in the buffer memory.The method of extracting frame amplitude spectrum is following:
At first the digitized sound signal of this frame is done pre-service (according to system's actual conditions, can comprise windowing, pre-emphasis etc.), establishing every frame length is the F point, and first zero padding is to N point (N>=F wherein, N=2
j, j is integer and j>=8), carry out leaf transformation in the N point discrete Fourier, obtain discrete spectrum
Y wherein
L, nN sampled point of l frame in the expression buffer memory, Y
L, kK Fourier transform value of i frame in the expression buffer memory (k=0,1 ..., N-1).So; It is that windowed function is (like Hanning window that its range value may be calculated
b (r); Breathe out peaceful window or the like), notice that the k here is omitted in the following description.
The initialization of step 200:HMM.Hidden Markov model λ of initialization on each frequency
M, wherein subscript M representes initialized time window length, and initialization procedure adopts constraint EM algorithm, and on certain frequency, concrete initialization step is following:
Step 201: the method through cluster (for example the non-supervision cluster of LBG, perhaps fuzzy clustering or the like) is divided into two types with M sample:
With
M wherein
0+ M
1=M, average bigger a type is with subscript (1) expression, another kind ofly representes with subscript (0).Two types average does
One type less average of energy does
Wherein
Two types variance is respectively:
Two types initializes weights coefficient:
The likelihood score of novel model of calculating,
In following iterative process, old model parameter set is expressed as λ '
M, new model parameter is:
Before the beginning iteration,
L ' is set to very big negative, for example a L '
k=-10000.The initialization forward direction factor; To the factor, begin interative computation below
after
initialization.
Step 202: calculate the forward direction factor:
Step 203: calculate back to the factor:
Step 204: calculating noise and voice probability of occurrence:
Step 206: calculate transition probability:
Step 207: calculate new initialization probability
Step 208: calculate new average
Step 209: the average to new retrains: μ '
1, k=max{ μ '
0, k, μ '
0, k+ δ }, wherein δ is a constant, span is between 0 to 100.
Step 210: calculate new variance
Step 211: new variance is retrained κ '
1, M=max{ κ '
0, M, κ '
1, M}
Step 212: calculate new transition probability,
Step 213: the likelihood score of novel model of calculating
Step 214: if satisfy condition
termination of iterations; Wherein ε is a very little numeral, for example ε=0.1.If
iteration jumps to " step 202 ".
The sequential renewal of step 300:HMM.Setting up initialized model λ
MAfterwards, since the M+1 frame, adopt the method for incremental learning, by frame update HMM model.Iterative process can be expressed as: on each frequency, and known λ
lWith current observed value x
l, infer λ
L+1Carry out Fourier transform for the l+1 frame, obtain T
L+1, k, 0≤k<N wherein.On each frequency; Calculate range value
for each frequency, following in the parameter update step of l+1 frame:
Step 301: calculate the forward direction factor,
Step 302: computing voice and noise probability of occurrence,
Step 303: the design conditions transition probability,
Step 304: calculating average noise voice probability of occurrence,
Step 305: rely on smoothing factor computing time,
Step 306: computation of mean values,
Step 307: the average to new retrains: μ
1, l+1=max{ μ
1, l+1, μ
0, l+1+ δ }.
Step 308: calculate new variance,
Step 309: new variance is retrained κ
1, l+1=max{ κ
0, l+1, κ
1, l+1}
Step 310: calculate average transition probability,
Step 311: calculate transition probability,
Step 312: the transition probability to new retrains, a
01, l=max{a
01, l, η }, a
00, l=1-a
01, l, a
10, l=max{a
10, l, η }, a
11, l=1-a
10, l
From above substep, we have obtained λ
L+1In all parameters, thereby obtained the relevant voice probability of occurrence
Power spectrum valuation μ with noise signal
0, l+1
Algorithm based on the foregoing description; The noise power spectrum estimation performance is estimated; Adopt each 8 sentence of men and women words person speech data and white Gaussian noise, F16 fight support storehouse noise and babble noise in the NOISEX92 noise data storehouse in the TIMIT database according to 0,5, signal to noise ratio (S/N ratio) such as 10dB mixes.First kind of evaluation index linear segmented error defines as follows:
D (k wherein; L) the actual noise amplitude spectrum of expression; The noise amplitude spectrum that
expression is estimated; Notice that error amount is more little, the expression estimated value approaches actual value more, and it is accurate more to estimate.Second kind of evaluation index logarithm segmentation error defines as follows:
In like manner, error amount is more little, and the expression estimating noise is accurate more.
Algorithm compares with three kinds of noise power spectrum algorithm for estimating of current main-stream respectively; Wherein MS representes the minimum statistics algorithm; MCRA representes the recurrence average algorithm of minimum control, and IMCRA representes that the minimum control that improves version returns average algorithm, and SHMM is an algorithm of the present invention.Table 1 has been expressed the result of line spectrum error SegError.
Noise Estimation linear segmented error under the various environment of table 1
Noise Estimation logarithm segmentation error under the various environment of table 2
Can find out that from last table the algorithm that the present invention proposes all has remarkable advantages for three kinds of algorithms of present main flow.
In addition, Fig. 2 has compared SHMM Noise Estimation algorithm, classical minimum statistics algorithm (MS), minimum control recurrence average algorithm (MCRA) and its raising version IMCRA through an instance.In this example, Noisy Speech Signal was 3.75 seconds position, and signal to noise ratio (S/N ratio) drops to 4dB suddenly from 10dB, and the position at 13.1 seconds rises to 10dB from 4dB again.(a) be the power spectrum of noisy speech on a certain subband; (b) the noise power spectrum envelope of estimating for the MS algorithm, wherein dotted line is represented real noise power spectrum envelope; (c) estimated result of expression MCRA algorithm; (d) estimated result of expression IMCRA algorithm; (e) represent the estimated result of this algorithm.As can be seen from the figure, other three algorithms are slower for the reaction of the unexpected rising of noise, can not catch up with jumping of noise fast.And the better performances that the SHMM algorithm table reveals.
In decades in the past, people have invented various algorithms, are used to estimate voice activity and noise power spectrum.Voice signal is one of them important clue in the sequential correlativity of frequency domain, because voice signal is non-stabilization signal, this sequential correlativity is also along with the time changes.Yet the algorithm in past does not cause enough attention to this sequential correlativity, just is used simply, becomes correlativity when not adopting adaptive method to describe.The present invention adopts sequential hidden Markov model SHMM to describe the sequential correlativity of voice on each frequency component; This sequential estimation method is to be based upon on the basis of single order recurrence; Sequential correlativity and its parameter set change along with the variation of input signal together.This statistical model has accurately reflected the sequential statistical nature of voice, and therefore, the algorithm for estimating that the present invention proposes is superior to the algorithm (for example minimum statistics, the recurrence of minimum control is average) of present main flow in performance.
In addition, the SHMM model all was on the basis that is based upon the high-order recurrence in the past, and the single order that proposes among the present invention returns SHMM with respect to the SHMM that high-order returns, and has practiced thrift calculated amount and storage space greatly.This is another innovation part of the present invention.
It should be noted last that above embodiment is only unrestricted in order to technical scheme of the present invention to be described.Although the present invention is specified with reference to embodiment; Those of ordinary skill in the art is to be understood that; Technical scheme of the present invention is made amendment or is equal to replacement, do not break away from the spirit and the scope of technical scheme of the present invention, it all should be encompassed in the middle of the claim scope of the present invention.
Claims (10)
1. a noise power spectrum is estimated and the voice activity detection method; This method is described the sequential correlativity of voice on each frequency component based on the sequential hidden Markov model SHMM that single order returns; And adopt the mode of incremental learning to come progressively to upgrade SHMM, and final, deduce out the power spectrum information of the probability of occurrence on this frequency subband and the noise of voice; With the sequential statistical nature of accurate reflection voice, this method comprises the following steps:
1) on each frequency component, extracts logarithm amplitude spectrum envelope for voice signal; And set up the binary hidden Markov model of a correspondence, wherein, one-component is represented the distribution of speech energy; Another component is the distribution of noise energy, and each state is represented by Gaussian distribution;
2) for one section speech data, set the M frame buffer, deposit preceding M frame input signal in the buffer memory in, extract the logarithm amplitude spectrum of M frame in the buffer memory, adopt the maximum likelihood algorithm for estimating to set up an initialized model;
3) obtaining initialized model λ
MAfterwards, since the M+1 frame, adopt the method for incremental learning, by the HMM model of each frequency band of frame update, recursion obtains the probability of occurrence of noise figure and voice signal successively.
2. noise power spectrum according to claim 1 estimates and the voice activity detection method that the concrete steps of this method comprise:
1) on each frequency component, extracts logarithm amplitude spectrum envelope for voice signal, for the logarithm amplitude spectrum time series x on the frequency component
l={ x
1, x
2..., x
l, set up a hidden Markov model s
l={ s
1, s
2..., s
l, s
t{ 0,1} is its corresponding status switch to ∈, and 1 expression voice go out present condition, and 0 expression noise goes out present condition, λ
lExpression is from sequence x
lIn the model parameter valuation of obtaining, so, for a given parameter set λ
l, corresponding observed value sequence x
lProbability density function can be expressed as:
Wherein, p (s
l| λ
l) expression status switch s
lThe prior probability that occurs, gaussian component is expressed as:
Here
The expression state transition probability,
Expression original state probability, p (x
l| λ
l, s
l) expression given state s
lWith parameter set λ
lSituation under observed value sequence x
lLikelihood score:
Wherein,
changes;
μ in this model
0, lBe exactly that we want the noise estimated, simultaneously, we can derive that voice signal occurs on certain frequency of l frame probability does
2) for one section speech data, set the M frame buffer, deposit preceding M frame input signal in the buffer memory in, extract the logarithm amplitude spectrum of M frame in the buffer memory, the HMM model of substitution step 1) is to hidden Markov model λ of initialization on each frequency
M, subscript M representes initialized time window length, l>=M;
3) obtaining initialized model λ
MAfterwards, since the M+1 frame, the HMM model adopts the method for incremental learning, and by frame update SHMM model, recursion obtains λ successively
lAnd draw noise figure μ
0, lWith the probability of occurrence of voice signal on certain frequency of l frame.
3. noise power spectrum according to claim 1 and 2 is estimated and the voice activity detection method, it is characterized in that, the step of extracting frame amplitude spectrum in the described step 1) comprises:
At first, the digitized sound signal of this frame is done pre-service, establishing every frame length is the F point, and first zero padding is to N point, N>=F, N=2
j, j is an integer, and j>=8, carries out leaf transformation in the N point discrete Fourier, obtains discrete spectrum
Wherein, y
L, nN sampled point of l frame in the expression buffer memory, Y
L, kK Fourier transform value of i frame in the expression buffer memory (k=0,1 ..., N-1); So, its range value may be calculated
In the formula, b (r) is a windowed function.
4. noise power spectrum according to claim 3 is estimated and the voice activity detection method, it is characterized in that described pre-service comprises windowing or/and pre-emphasis.
5. noise power spectrum according to claim 3 is estimated and the voice activity detection method, it is characterized in that, described windowed function adopts Hanning window or breathes out peaceful window.
6. noise power spectrum according to claim 1 and 2 is estimated and the voice activity detection method, it is characterized in that described step 2) in the initialization of HMM, concrete initialized step comprises on certain frequency:
Step 201): the method through cluster is divided into two types with M sample:
With
Wherein, M
0+ M
1=M, average bigger a type is with subscript (1) expression, another kind ofly representes with subscript (0);
Two types variance is respectively:
Two types initializes weights coefficient is:
The likelihood score of novel model of calculating,
And beginning interative computation; In following iterative process, old model parameter set is expressed as λ '
M, new model parameter is:
Before the beginning iteration,
L ' is set to a very big negative, the initialization forward direction factor,
After the initialization to the factor,
Step 202): calculate the forward direction factor:
Step 203): calculate back to the factor:
Step 204): calculating noise and voice probability of occurrence:
Step 205): if
stops to fall generation, wherein ζ approaches zero but greater than zero decimal;
Step 206): calculate transition probability:
Step 207): calculate new initialization probability
Step 208): calculate new average
Step 209): the average to new retrains: μ '
1, M=max{ μ '
0, M, μ '
0, M+ δ }, wherein, δ is a constant, span is between 0 to 100;
Step 210): calculate new variance
Step 211): new variance is retrained κ '
1, M=max{ κ '
0, M, κ '
1, M;
Step 212): calculate new transition probability,
Step 213): the likelihood score of novel model of calculating
7. noise power spectrum according to claim 6 is estimated and the voice activity detection method, it is characterized in that described step 201) in the method for cluster adopt non-supervision cluster of LBG or fuzzy clustering method.
8. noise power spectrum according to claim 1 and 2 is estimated and the voice activity detection method, it is characterized in that the sequential renewal of the HMM in the described step 3) is to set up initialized model λ
MAfterwards, since the M+1 frame, adopt the method for incremental learning, by frame update HMM model, its iterative process can be expressed as: on each frequency, and known λ
lWith current observed value x
l, infer λ
L+1Carry out Fourier transform for the l+1 frame, obtain Y
L+1k, wherein, 0≤k<N; On each frequency, calculate range value
For each frequency, following in the parameter update step of l+1 frame:
Step 301): calculate the forward direction factor,
Z ∈ 0,1}.
Step 302): computing voice and noise probability of occurrence,
Step 303): the design conditions transition probability,
Step 304): calculating average noise voice probability of occurrence,
Step 305): rely on smoothing factor computing time,
Step 306): the computing mode average,
Step 307): the state average to new retrains: μ
1, l+1=max{ μ
1, l+1, μ
0, l+1+ δ }, l>=M;
Step 308): calculate new state variance,
Step 309): the new state variance is retrained κ
1, l+1=max{ κ
0, l+1, κ
1, l+1, l>=M;
Step 310): calculate average transition probability,
Step 311): the computing mode probability,
Step 312): the transition probability to new retrains, a
01, l=max{a
01, l, η }, a
00, l=1-a
01, l, a
10, l=max{a
10, l, η }, a
11, l=1-a
10, l, l>=M;
From above substep, obtained λ
L+1In all parameters, thereby obtained relevant voice probability of occurrence γ
L+1 λ l(1) and the power spectrum valuation μ of noise signal
0, l+1
9. noise power spectrum according to claim 8 is estimated and the voice activity detection method, it is characterized in that, the incremental learning method that the HMM model in the described step 3) adopts comprises: recursion weight coefficient, recursion average and pass the vertebra variance;
Wherein, Described recursion average: in
formula;
is a smoothing factor that depends on the voice probability of occurrence, less than 1 but approach 1;
Described recursion variance:
10. noise power spectrum according to claim 1 is estimated and the voice activity detection method, it is characterized in that, the parameter recurrence method of the described sequential hidden Markov model that returns based on single order is:
Calculate the forward direction factor of HMM:
Computing voice and noise probability of occurrence,
The design conditions transition probability,
Calculating average noise voice probability of occurrence,
Rely on smoothing factor computing time,
Computation of mean values,
Calculate new variance,
Calculate average transition probability,
Calculate transition probability,
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201110141137.5A CN102800322B (en) | 2011-05-27 | 2011-05-27 | Method for estimating noise power spectrum and voice activity |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201110141137.5A CN102800322B (en) | 2011-05-27 | 2011-05-27 | Method for estimating noise power spectrum and voice activity |
Publications (2)
Publication Number | Publication Date |
---|---|
CN102800322A true CN102800322A (en) | 2012-11-28 |
CN102800322B CN102800322B (en) | 2014-03-26 |
Family
ID=47199411
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201110141137.5A Expired - Fee Related CN102800322B (en) | 2011-05-27 | 2011-05-27 | Method for estimating noise power spectrum and voice activity |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN102800322B (en) |
Cited By (16)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103489454A (en) * | 2013-09-22 | 2014-01-01 | 浙江大学 | Voice endpoint detection method based on waveform morphological characteristic clustering |
CN104575513A (en) * | 2013-10-24 | 2015-04-29 | 展讯通信(上海)有限公司 | Burst noise processing system and burst noise detection and suppression method and device |
WO2015078268A1 (en) * | 2013-11-27 | 2015-06-04 | Tencent Technology (Shenzhen) Company Limited | Method, apparatus and server for processing noisy speech |
CN105355199A (en) * | 2015-10-20 | 2016-02-24 | 河海大学 | Model combination type speech recognition method based on GMM (Gaussian mixture model) noise estimation |
CN105810201A (en) * | 2014-12-31 | 2016-07-27 | 展讯通信(上海)有限公司 | Voice activity detection method and system |
CN106331969A (en) * | 2015-07-01 | 2017-01-11 | 奥迪康有限公司 | Enhancement of noisy speech based on statistical speech and noise models |
WO2017063516A1 (en) * | 2015-10-13 | 2017-04-20 | 阿里巴巴集团控股有限公司 | Method of determining noise signal, and method and device for audio noise removal |
CN106653047A (en) * | 2016-12-16 | 2017-05-10 | 广州视源电子科技股份有限公司 | Automatic gain control method and device for audio data |
CN107731230A (en) * | 2017-11-10 | 2018-02-23 | 北京联华博创科技有限公司 | A kind of court's trial writing-record system and method |
CN104269180B (en) * | 2014-09-29 | 2018-04-13 | 华南理工大学 | A kind of quasi- clean speech building method for speech quality objective assessment |
CN108113646A (en) * | 2016-11-28 | 2018-06-05 | 中国科学院声学研究所 | A kind of detection in cardiechema signals cycle and the state dividing method of heart sound |
CN108986832A (en) * | 2018-07-12 | 2018-12-11 | 北京大学深圳研究生院 | Ears speech dereverberation method and device based on voice probability of occurrence and consistency |
CN109616139A (en) * | 2018-12-25 | 2019-04-12 | 平安科技(深圳)有限公司 | Pronunciation signal noise power spectral density estimation method and device |
CN109741759A (en) * | 2018-12-21 | 2019-05-10 | 南京理工大学 | A kind of acoustics automatic testing method towards specific birds species |
CN110136738A (en) * | 2019-06-13 | 2019-08-16 | 苏州思必驰信息科技有限公司 | Noise estimation method and device |
WO2021057239A1 (en) * | 2019-09-23 | 2021-04-01 | 腾讯科技(深圳)有限公司 | Speech data processing method and apparatus, electronic device and readable storage medium |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1242553A (en) * | 1998-03-24 | 2000-01-26 | 松下电器产业株式会社 | Speech detection system for noisy conditions |
US20020188445A1 (en) * | 2001-06-01 | 2002-12-12 | Dunling Li | Background noise estimation method for an improved G.729 annex B compliant voice activity detection circuit |
CN101853661A (en) * | 2010-05-14 | 2010-10-06 | 中国科学院声学研究所 | Noise spectrum estimation and voice mobility detection method based on unsupervised learning |
CN102044243A (en) * | 2009-10-15 | 2011-05-04 | 华为技术有限公司 | Method and device for voice activity detection (VAD) and encoder |
-
2011
- 2011-05-27 CN CN201110141137.5A patent/CN102800322B/en not_active Expired - Fee Related
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1242553A (en) * | 1998-03-24 | 2000-01-26 | 松下电器产业株式会社 | Speech detection system for noisy conditions |
US20020188445A1 (en) * | 2001-06-01 | 2002-12-12 | Dunling Li | Background noise estimation method for an improved G.729 annex B compliant voice activity detection circuit |
CN102044243A (en) * | 2009-10-15 | 2011-05-04 | 华为技术有限公司 | Method and device for voice activity detection (VAD) and encoder |
CN101853661A (en) * | 2010-05-14 | 2010-10-06 | 中国科学院声学研究所 | Noise spectrum estimation and voice mobility detection method based on unsupervised learning |
Non-Patent Citations (1)
Title |
---|
赵力: "《语音信号处理》", 31 March 2003, 机械工业出版社 * |
Cited By (25)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103489454B (en) * | 2013-09-22 | 2016-01-20 | 浙江大学 | Based on the sound end detecting method of wave configuration feature cluster |
CN103489454A (en) * | 2013-09-22 | 2014-01-01 | 浙江大学 | Voice endpoint detection method based on waveform morphological characteristic clustering |
CN104575513B (en) * | 2013-10-24 | 2017-11-21 | 展讯通信(上海)有限公司 | The processing system of burst noise, the detection of burst noise and suppressing method and device |
CN104575513A (en) * | 2013-10-24 | 2015-04-29 | 展讯通信(上海)有限公司 | Burst noise processing system and burst noise detection and suppression method and device |
WO2015078268A1 (en) * | 2013-11-27 | 2015-06-04 | Tencent Technology (Shenzhen) Company Limited | Method, apparatus and server for processing noisy speech |
US9978391B2 (en) | 2013-11-27 | 2018-05-22 | Tencent Technology (Shenzhen) Company Limited | Method, apparatus and server for processing noisy speech |
CN104269180B (en) * | 2014-09-29 | 2018-04-13 | 华南理工大学 | A kind of quasi- clean speech building method for speech quality objective assessment |
CN105810201B (en) * | 2014-12-31 | 2019-07-02 | 展讯通信(上海)有限公司 | Voice activity detection method and its system |
CN105810201A (en) * | 2014-12-31 | 2016-07-27 | 展讯通信(上海)有限公司 | Voice activity detection method and system |
CN106331969A (en) * | 2015-07-01 | 2017-01-11 | 奥迪康有限公司 | Enhancement of noisy speech based on statistical speech and noise models |
WO2017063516A1 (en) * | 2015-10-13 | 2017-04-20 | 阿里巴巴集团控股有限公司 | Method of determining noise signal, and method and device for audio noise removal |
US10796713B2 (en) | 2015-10-13 | 2020-10-06 | Alibaba Group Holding Limited | Identification of noise signal for voice denoising device |
CN105355199B (en) * | 2015-10-20 | 2019-03-12 | 河海大学 | A kind of model combination audio recognition method based on the estimation of GMM noise |
CN105355199A (en) * | 2015-10-20 | 2016-02-24 | 河海大学 | Model combination type speech recognition method based on GMM (Gaussian mixture model) noise estimation |
CN108113646A (en) * | 2016-11-28 | 2018-06-05 | 中国科学院声学研究所 | A kind of detection in cardiechema signals cycle and the state dividing method of heart sound |
CN106653047A (en) * | 2016-12-16 | 2017-05-10 | 广州视源电子科技股份有限公司 | Automatic gain control method and device for audio data |
CN107731230A (en) * | 2017-11-10 | 2018-02-23 | 北京联华博创科技有限公司 | A kind of court's trial writing-record system and method |
CN108986832A (en) * | 2018-07-12 | 2018-12-11 | 北京大学深圳研究生院 | Ears speech dereverberation method and device based on voice probability of occurrence and consistency |
CN108986832B (en) * | 2018-07-12 | 2020-12-15 | 北京大学深圳研究生院 | Binaural voice dereverberation method and device based on voice occurrence probability and consistency |
CN109741759B (en) * | 2018-12-21 | 2020-07-31 | 南京理工大学 | Acoustic automatic detection method for specific bird species |
CN109741759A (en) * | 2018-12-21 | 2019-05-10 | 南京理工大学 | A kind of acoustics automatic testing method towards specific birds species |
CN109616139A (en) * | 2018-12-25 | 2019-04-12 | 平安科技(深圳)有限公司 | Pronunciation signal noise power spectral density estimation method and device |
CN109616139B (en) * | 2018-12-25 | 2023-11-03 | 平安科技(深圳)有限公司 | Speech signal noise power spectral density estimation method and device |
CN110136738A (en) * | 2019-06-13 | 2019-08-16 | 苏州思必驰信息科技有限公司 | Noise estimation method and device |
WO2021057239A1 (en) * | 2019-09-23 | 2021-04-01 | 腾讯科技(深圳)有限公司 | Speech data processing method and apparatus, electronic device and readable storage medium |
Also Published As
Publication number | Publication date |
---|---|
CN102800322B (en) | 2014-03-26 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN102800322B (en) | Method for estimating noise power spectrum and voice activity | |
CN100543842C (en) | Realize the method that ground unrest suppresses based on multiple statistics model and least mean-square error | |
CN102800316B (en) | Optimal codebook design method for voiceprint recognition system based on nerve network | |
CN109272990A (en) | Audio recognition method based on convolutional neural networks | |
Cui et al. | Noise robust speech recognition using feature compensation based on polynomial regression of utterance SNR | |
CN109192200B (en) | Speech recognition method | |
CN101853661A (en) | Noise spectrum estimation and voice mobility detection method based on unsupervised learning | |
JPS62231996A (en) | Allowance evaluation of word corresponding to voice input | |
CN103345923A (en) | Sparse representation based short-voice speaker recognition method | |
CN105139864A (en) | Voice recognition method and voice recognition device | |
Frey et al. | Algonquin-learning dynamic noise models from noisy speech for robust speech recognition | |
Todkar et al. | Speaker recognition techniques: A review | |
Jung et al. | Linear-scale filterbank for deep neural network-based voice activity detection | |
CN103345920A (en) | Self-adaptation interpolation weighted spectrum model voice conversion and reconstructing method based on Mel-KSVD sparse representation | |
CN111081273A (en) | Voice emotion recognition method based on glottal wave signal feature extraction | |
Liang et al. | An improved noise-robust voice activity detector based on hidden semi-Markov models | |
Li et al. | A Convolutional Neural Network with Non-Local Module for Speech Enhancement. | |
Wan et al. | Variational bayesian learning for removal of sparse impulsive noise from speech signals | |
Pham et al. | Using artificial neural network for robust voice activity detection under adverse conditions | |
CN116189671A (en) | Data mining method and system for language teaching | |
Arslan et al. | Noise robust voice activity detection based on multi-layer feed-forward neural network | |
Van Dalen | Statistical models for noise-robust speech recognition | |
Xiao et al. | Single-channel speech separation method based on attention mechanism | |
Kathania et al. | Soft-weighting technique for robust children speech recognition under mismatched condition | |
Chen et al. | Machine Learning for Predictive Analytics in the Improvement of English Speech Feature Recognition |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20140326 |