CN108231087A - A kind of single channel blind source separating method - Google Patents
A kind of single channel blind source separating method Download PDFInfo
- Publication number
- CN108231087A CN108231087A CN201711393096.2A CN201711393096A CN108231087A CN 108231087 A CN108231087 A CN 108231087A CN 201711393096 A CN201711393096 A CN 201711393096A CN 108231087 A CN108231087 A CN 108231087A
- Authority
- CN
- China
- Prior art keywords
- signal
- single channel
- matrix
- imf
- point
- 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 24
- 230000000694 effects Effects 0.000 claims abstract description 20
- 238000000354 decomposition reaction Methods 0.000 claims abstract description 18
- 230000009467 reduction Effects 0.000 claims abstract description 10
- 239000000654 additive Substances 0.000 claims abstract description 4
- 230000000996 additive effect Effects 0.000 claims abstract description 4
- 239000011159 matrix material Substances 0.000 claims description 25
- 238000012545 processing Methods 0.000 claims description 16
- 230000010355 oscillation Effects 0.000 claims description 7
- 108010076504 Protein Sorting Signals Proteins 0.000 claims description 4
- 238000004458 analytical method Methods 0.000 claims description 4
- 238000007781 pre-processing Methods 0.000 claims description 4
- 239000000203 mixture Substances 0.000 claims description 3
- 238000010606 normalization Methods 0.000 claims description 3
- 238000012887 quadratic function Methods 0.000 claims description 3
- 238000000926 separation method Methods 0.000 claims description 3
- 230000017105 transposition Effects 0.000 claims description 3
- 238000010276 construction Methods 0.000 claims description 2
- 238000001228 spectrum Methods 0.000 abstract description 6
- 238000000513 principal component analysis Methods 0.000 abstract description 2
- 238000012880 independent component analysis Methods 0.000 abstract 1
- 230000006870 function Effects 0.000 description 6
- 238000002474 experimental method Methods 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 238000012216 screening Methods 0.000 description 2
- 230000002950 deficient Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 230000005764 inhibitory process Effects 0.000 description 1
- 230000005236 sound signal Effects 0.000 description 1
- 238000010183 spectrum analysis Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L21/00—Speech or voice signal processing techniques 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
- G10L21/0272—Voice signal separating
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
- G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L21/00—Speech or voice signal processing techniques to produce another audible or non-audible signal, e.g. visual or tactile, in order to modify its quality or its intelligibility
- G10L21/02—Speech enhancement, e.g. noise reduction or echo cancellation
- G10L21/0208—Noise filtering
- G10L2021/02087—Noise filtering the noise being separate speech, e.g. cocktail party
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- General Engineering & Computer Science (AREA)
- Quality & Reliability (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computational Linguistics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Signal Processing (AREA)
- Health & Medical Sciences (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Human Computer Interaction (AREA)
- Acoustics & Sound (AREA)
- Multimedia (AREA)
- Noise Elimination (AREA)
Abstract
A kind of single channel blind source separating method, belongs to electronic information technical field, is characterized in the method using extreme value point-symmetric extension, carries out end effect to overall experience mode decomposition and handle；And single channel mixed signal is converted into intrinsic mode function (IMFs), and inhibit noise with the overall experience Mode Decomposition；Dimension-reduction treatment is carried out to multichannel IMFs using principal component analysis, removes invalid components therein；Multiple signals after dimensionality reduction are subjected to independent component analysis to realize blind source separating.Implementation steps are that multiple signals linear, additive is mixed into single channel signal to be transmitted, simple, quick, effectively recover source signal finally under the conditions of late mode recognition effect is not influenced, and realize the output of multiple-channel output mouth.Advantage is to separate the signal mixed as multichannel spectrum overlapping all the way in the case where not influencing later stage recognition effect.
Description
Technical field
The invention belongs to electronic information technical fields, and in particular to a kind of single channel blind source separating method.
Background technology
" cocktail party problem " that the most classical example application of blind source separating is known as, this problem are to be based on being in this way
One scene：In the cocktail party participated in many people, everybody is talking, and various sound mix,
Assuming that we with microphone records these voice signals, require us to be detached from numerous sound mix signal together now
Obtain the voice of someone, due to people is relatively more and the limited amount of microphone, the problem of this has reformed into deficient positive definite.Wish
There is the signal that a kind of multichannel is mixed into all the way to be detached, can effectively recover multichannel source beginning signal, can thus obtain
The recording talked to your interested people, the single channel blind source separating method of the prior art have following three types：1. single channel ICA
Analysis, when the frequency spectrum of signal is at a distance of relatively closely, such as mixed signal for mother and baby's heartbeat cannot be detached with the method；It is 2. right
Signal singular values carry out ICA processing after carrying out ICA processing and singular spectrum analysis after decomposing again, this two methods is for signal frequency
During spectrum overlapping, separation signal effect is poor, aliasing occurs；3. ICA processing, i.e. W_ICA and empirical modal are carried out after wavelet decomposition
ICA processing, i.e. EMD_ICA are carried out after decomposition, this two methods can still provide for detaching in the case of the spectrum overlapping of signal, transport
It needs to carry out selection small echo for different signals during with wavelet decomposition, and empirical mode decomposition is the feature extraction according to signal
Go out intrinsic mode function (Intrinsic Model Function), i.e. IMF has very strong adaptivity；By being acquired
To some signal spectrums can be overlapped in any case, it is practical to compare W_ICA, EMD_ICA and EEMD_ICA, find EMD_
ICA separating effect waveforms are smooth, and closer to original signal, but the method processing procedure speed is slower, need people in the process by warp
Test and carry out signal and select, it is intelligent not high, and EEMD_ICA is better than EMD_ICA in terms of noise is inhibited, but with EMD_ICA mono-
There is end effect in sample.During " screening " of intrinsic mode function, the cubic spline letter of envelope up and down is formed
Number will appear Divergent Phenomenon at the both ends of data sequence, and the result of this diverging can with " screening " process it is continuous into
Row, gradual inwardly " pollution " entire data sequence, and make obtained result serious distortion.
Invention content
It is an object of the present invention to provide a kind of single channel blind source separating method, the digital signal being mixed into multichannel all the way is divided
From can effectively recover multichannel source beginning signal.
The invention is realized in this way specific implementation step is：
A, multiple signals will be collected in signal pre-processing module linear, additive, obtain pretreatment single channel signal x (t)；
B, obtained single channel signal x (t) will be pre-processed and be sent to signal blind source separating module, carried out at end effect
Reason, using the method for extreme value point-symmetric extension (Extreme point symmetry extension, EPSE), Ran Houyi
It is secondary progress overall experience mode decomposition (Ensemble Empirical Mode Decomposition), i.e. EEMD decompose, it is main into
Part analysis PCA dimensionality reductions and ICA analyses, realize that multiple signals are acquired by an input port, the output of multiple-channel output mouth：
B.1, inhibit extreme value point-symmetric extension (EPSE) algorithm of end effect：
A, the signal x (t) obtained to pretreatment is N's to length to external symmetry plus extreme point using endpoint as symmetric points
Discrete signal sequence：X (i), T (i)=i, i=1,2 ..., N, very big value sequence are：U (i), Tu (i), i=1,2 ...,
Nu, wherein U (i)=S (Tu (i)), minimum ordered series of numbers are：L (i), Tl (i), i=1,2 ..., Nl, wherein L (i)=S (Tl
(i)), at former data endpoint, using endpoint as symmetric points, it is symmetrically extended the extreme point in Nc period outward, if signal sequence
Periodicity be less than setting value, then Nc takes cycle value, and the extreme value sequence by continuation is：
B, the endpoint of former data sequence is most likely not extreme point, if using it as extreme point, will hold envelope
It is shunk at point, makes envelope shape gross distortion, introduce oscillation error, when endpoint value exceeds certain range, held to avoid
The drift phenomenon of point will be inserted into extreme value point sequence Ua, Tua, La, Tla above as extreme point, for the sake of simplicity, with
Benchmark of the extreme value point value of proximal points as judgement,
Forward terminal x (l)：
Aft terminal x (N)：
Period T (i)=i, i=1,2 ..., N is fitted using treated extreme value point sequence Ua, Tua, La, Tla
Obtain the envelope up and down of original signal；
B.2, overall experience mode decomposition EEMD processing obtains intrinsic mode function IMFs：
A, obtained single channel signal x (t) will be handled, it is zero repeatedly to add in mean value, standard deviation is the white of constant
Noise ni (t), as xi (t)=x (t)+ni (t), wherein xi (t) are the signal added in after white noise, and ni (t) is ith
The white noise of addition, the criterion of white noise ni (t) is wherein, ε n represent the white Gaussian noise standard deviation added in, ε h expression signals
In effectively radio-frequency component amplitude standard deviation, ε 0 represent signal amplitude standard deviation, α is proportionality coefficient, it is generally the case that α=σ/4
The modal aliasing in signal decomposition can effectively be avoided；
B, empirical mode decomposition EMD is carried out to obtained signal, obtains respective IMF and be denoted as a ij (t) and remainder ri
(t), wherein a ij (t) represent to add in j-th of the IMF decomposed after white noise；
C, the obtained IMF of step b are subjected to population mean operation, obtain IMFs as wherein, aj (t) be to original signal into
Obtained j-th of IMF after row EEMD is decomposed, it is hereby achieved that IMF Component Matrices A={ a1 (t), a2 (t) ..., ai (t) }
T, wherein, i is IMF number of the single channel signal after EEMD is decomposed, and subscript T is transposition operation；
B.3 PCA dimensionality reductions, are carried out to obtaining IMF components：R=E (AA T), wherein RV=V Λ, A are after EEMD is decomposed
Obtained IMF component m * n matrixes, R are the autocorrelation matrix of m variable IMFs, and V is m × m rank eigenvectors matrixs of R,
Column vector is the feature vector of the orthogonal normalization of R, and Λ is the feature diagonal matrix of R, and λ i (i=1,2 ..., m) are i-th pair
Element on linea angulata, construction m incoherent new variables Y=VTX, Y={ y1, y2 ..., ym } T, to λ i (i=1,2 ..., m)
After arranging in descending order, the feature vector corresponding to front p larger characteristic values is taken, obtains p × n rank vector matrix B after dimensionality reduction,
Middle p >=2；
B.4 matrix B obtained by PCA dimensionality reductions, is subjected to ICA processing, good using effect and fireballing basic fixed point
Iterative algorithm FastICA carries out ICA processing：A, whitened data, provides observing matrix X, wherein X=AS (n), and A is the mixed of signal
Matrix is closed, S (n) is source signal；B, weights of the weight vector as the mixed matrix W of solution are randomly selected, W is the mixed matrix of solution, is as mixed
The pseudo inverse matrix of matrix A；C, W ← E (xg (WTx))-E (g ' (WTx)) W is enabled, wherein the derivative g (u) of non-quadratic function G=
Tanh (alu), 1≤a1≤2 are constants, often take and do 1；D, loop iteration until convergence, finally obtains p × n ranks vector
Matrix Y (n), by observing the signal after choosing separation.
Advantage of the present invention and good effect：Letter for multichannel spectrum overlapping all the way can will be mixed using the method for the present invention
Number, it is separated in the case where not influencing later stage recognition effect.Such as in international conference, many people participate in discussing issues,
In the case that number of microphone is limited, to accurately find the sound of someone in the recording of numerous people, using the present invention
Method can reach good effect.
Description of the drawings
Fig. 1 is the block diagram of model of the present invention；
Fig. 2 is the flow chart of model of the present invention；
Fig. 3 is the two-way source signal and mixed signal that experiment uses, and s1 (t) is chooses from the recording of a female announcer
The voice signal elected, s2 (t) are the oscillation mode signal as noise section, and x (t) is mixed single channel signal；
Fig. 4 is the signal after restoring, and s1* (t) is the voice signal of female announcer recovered, and s2* (t) is recovered
Oscillation mode signal as noise section.
Specific embodiment
Implemented for female voice signal and analyzed by being isolated from the source signal of an oscillation mode, step is such as
Under：
1st, the two paths of signals of experiment, if Fig. 3 is the select voice letter from the recording of a female announcer of s1 (t)
Number, sample frequency 8KHz, s2 (t) they are the oscillation mode signal as noise section, be the sinusoidal signal that is generated with matlab come
It represents, signal pre-processing module is transferred to by shielded wire, carry out linear, additive and be mixed to get output signal, such as Fig. 3 x
(t)。
2nd, it to the output signal after signal pre-processing module, is normalized first, then carries out inhibition end
The EPSE algorithm process of point effect subsequently enters EEMD resolution process modules, extracts intrinsic mode function IMFs, is dropped into PCA
Dimension extracts pivot, finally carries out ICA processing.Multiple signals are isolated, if Fig. 4 s1* (t) are the language of female announcer that recovers
Sound signal, s2* (t) are the oscillation mode signal as noise section recovered.
In the present embodiment, in above-mentioned steps 2, the processing of end effect is inhibited to be as follows：
A, the signal x (t) obtained to pretreatment, using endpoint as symmetric points, to external symmetry plus extreme point.
B, the endpoint of former data sequence is most likely not extreme point, for the sake of simplicity, using the extreme value point value of proximal points as
The benchmark of judgement.At the endpoint of former data, using endpoint as symmetric points, it is symmetrically extended the extreme point in 8 periods outward, if letter
The periodicity of number sequence is less than setting value, then 8 take cycle value.
In the present embodiment, in above-mentioned steps 2, EEMD resolution process obtains intrinsic mode function IMFs；It is as follows：
A, will treated single channel signal x (t), adding in 100 times has the white noise that mean value is zero, standard deviation is constant
Ni (t), as xi (t)=x (t)+ni (t), wherein xi (t) are the signal added in after white noise, and ni (t) is added in for ith
White noise, the standard deviation of white noise is 0.2；
B, EMD decomposition is carried out to obtained signal, obtains respective IMF and be denoted as a ij (t) and remainder ri (t).Wherein
A ij (t) represent to add in j-th of the IMF decomposed after white noise；
C, obtained IMF carries out population mean operation, obtains IMFs as wherein, aj (t) is to carry out EEMD to original signal
Obtained j-th of IMF after decomposition.It is hereby achieved that IMF Component Matrices A={ a1 (t), a2 (t) ..., ai (t) } T,
In, i=12 is IMF number of the single channel signal after EEMD is decomposed, and subscript T is transposition operation；
In the present embodiment, in above-mentioned steps 2, principal component analysis PCA processing is as follows：R=E (AA T), RV=V
Λ, wherein A are 12 × 3000 matrix of IMF components obtained after EEMD is decomposed, and R is the auto-correlation square of 12 variable IMFs
Battle array, V are 12 × 12 rank eigenvectors matrixs of R, and column vector is the feature vector of the orthogonal normalization of R, and Λ is the feature pair of R
Angular moment battle array, λ i, i=1,2 ..., 12 be the element on i-th of diagonal, constructs 12 incoherent new variables Y=VTX, Y=
{ y1, y2 ..., y12 } T after being arranged in descending order λ i (i=1,2 ..., 12), takes the spy corresponding to the larger characteristic value in front 2
Sign vector, obtains 2 × 3000 vector matrix B after dimensionality reduction；
In this example, in above-mentioned steps 2, ICA processing using effect preferably, the basic fixed point iteration of speed calculates
Method FastICA.It is specific as follows to carry out FastICA processing：
A, whitened data provides observing matrix X；
B, weights of the weight vector as the mixed matrix W of solution are randomly selected；
C, W ← E (xg (WTx))-E (g ' (WTx)) W is enabled, wherein derivative g (u)=tanh (alu) of non-quadratic function G, a1
It takes and does 1；
D, loop iteration, until convergence.2 × 3000 rank vector matrix Y (n) are finally obtained, signal after as detaching.
Embodiment effect is to calculate Fig. 3 s1 (t) and Fig. 4 s1* (t) related coefficients as 0.7895, Fig. 3 s2 (t) and Fig. 4 s2*
(t) related coefficient is 0.9944, shows that the present invention can more actually recover it for the single channel blind source separating method of signal
Preceding signal.
Claims (1)
1. a kind of single channel blind source separating method, it is characterised in that implementation steps are：
A, multiple signals will be collected in signal pre-processing module linear, additive, obtain pretreatment single channel signal x (t)；
B, obtained single channel signal x (t) will be pre-processed and be sent to signal blind source separating module, end effect is carried out and handle,
Using the method for extreme value point-symmetric extension (Extreme point symmetry extension, EPSE), then successively
Overall experience mode decomposition (Ensemble Empirical Mode Decomposition) is carried out, i.e. EEMD is decomposed, main composition
PCA dimensionality reductions and ICA analyses are analyzed, realizes that multiple signals are acquired by an input port, the output of multiple-channel output mouth：
B.1, inhibit extreme value point-symmetric extension (EPSE) algorithm of end effect：
A, the signal x (t) obtained to pretreatment is the discrete of N to length to external symmetry plus extreme point using endpoint as symmetric points
Signal sequence：X (i), T (i)=i, i=1,2 ..., N, very big value sequence are：U (i), Tu (i), i=1,2 ..., Nu,
Middle U (i)=S (Tu (i)), minimum ordered series of numbers are：L (i), Tl (i), i=1,2 ..., Nl, wherein L (i)=S (Tl (i)),
At former data endpoint, using endpoint as symmetric points, it is symmetrically extended the extreme point in Nc period outward, if the period of signal sequence
Number is less than setting value, then Nc takes cycle value, and the extreme value sequence by continuation is：
B, the endpoint of former data sequence is most likely not extreme point, if using it as extreme point, will make envelope at endpoint
It shrinks, makes envelope shape gross distortion, introduce oscillation error, when endpoint value exceeds certain range, to avoid endpoint
Drift phenomenon will be inserted into extreme value point sequence Ua, Tua, La, Tla above, for the sake of simplicity, with proximal end as extreme point
Benchmark of the extreme value point value of point as judgement,
Period T (i)=i, i=1,2 ..., N are fitted to obtain using treated extreme value point sequence Ua, Tua, La, Tla
The envelope up and down of original signal；
B.2, overall experience mode decomposition EEMD processing obtains intrinsic mode function IMFs：
A, obtained single channel signal x (t) will be handled, will repeatedly be added in the white noise that mean value is zero, standard deviation is constant
Ni (t), as xi (t)=x (t)+ni (t), wherein xi (t) are the signal added in after white noise, and ni (t) is added in for ith
White noise, for wherein, ε n represent the white Gaussian noise standard deviation added in, ε h represent have in signal for the criterion of white noise ni (t)
The amplitude standard deviation of radio-frequency component is imitated, ε 0 represents signal amplitude standard deviation, and α is proportionality coefficient, it is generally the case that α=σ/4 can have
Effect avoids the modal aliasing in signal decomposition；
B, empirical mode decomposition EMD is carried out to obtained signal, obtains respective IMF and be denoted as aij (t) and remainder ri (t),
Wherein aij (t) represents to add in j-th of the IMF decomposed after white noise；
C, the obtained IMF of step b are subjected to population mean operation, obtain IMFs as wherein, aj (t) is that original signal is carried out
Obtained j-th of IMF after EEMD is decomposed, it is hereby achieved that IMF Component Matrices A={ a1 (t), a2 (t) ..., ai (t) } T,
Wherein, i is IMF number of the single channel signal after EEMD is decomposed, and subscript T is transposition operation；
B.3 PCA dimensionality reductions, are carried out to obtaining IMF components：R=E (AAT), wherein RV=V Λ, A is obtain after EEMD is decomposed
IMF component m * n matrixes, R is the autocorrelation matrix of m variable IMFs, m × m rank eigenvectors matrixs of the V for R, arrange to
Amount is the feature vector of the orthogonal normalization of R, and Λ is the feature diagonal matrix of R, and λ i (i=1,2 ..., m) are i-th of diagonal
On element, construction m incoherent new variables Y=VTX, Y={ y1, y2 ..., ym } T, to λ i (i=1,2 ..., m) by drop
After sequence arrangement, the feature vector corresponding to front p larger characteristic values is taken, obtains p × n ranks vector matrix B, wherein p after dimensionality reduction
≥2；
B.4 matrix B obtained by PCA dimensionality reductions, is subjected to ICA processing, good using effect and fireballing basic fixed point iteration
Algorithm FastICA carries out ICA processing：A, whitened data, provides observing matrix X, wherein X=AS (n), and A is the mixed moment of signal
Battle array, S (n) are source signal；B, weights of the weight vector as the mixed matrix W of solution are randomly selected, W mixes matrix, as hybrid matrix A for solution
Pseudo inverse matrix；C, W ← E (xg (WTx))-E (g ' (WTx)) W is enabled, wherein derivative g (u)=tanh of non-quadratic function G
(a1u), 1≤a1≤2, are constants, often take and do 1；D, loop iteration until convergence, finally obtains p × n rank vector matrixs Y
(n), by observing the signal after choosing separation.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711393096.2A CN108231087A (en) | 2017-12-14 | 2017-12-14 | A kind of single channel blind source separating method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711393096.2A CN108231087A (en) | 2017-12-14 | 2017-12-14 | A kind of single channel blind source separating method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN108231087A true CN108231087A (en) | 2018-06-29 |
Family
ID=62648344
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711393096.2A Pending CN108231087A (en) | 2017-12-14 | 2017-12-14 | A kind of single channel blind source separating method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108231087A (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108962276A (en) * | 2018-07-24 | 2018-12-07 | 北京三听科技有限公司 | A kind of speech separating method and device |
| CN111242366A (en) * | 2020-01-08 | 2020-06-05 | 广东技术师范大学 | EMD method and device for processing signals in real time |
| CN111246285A (en) * | 2020-03-24 | 2020-06-05 | 北京奇艺世纪科技有限公司 | Method for separating sound in comment video and method and device for adjusting volume |
| CN111784068A (en) * | 2020-07-09 | 2020-10-16 | 北京理工大学 | EEMD-based power load combined prediction method and device |
| CN113314137A (en) * | 2020-02-27 | 2021-08-27 | 东北大学秦皇岛分校 | Mixed signal separation method based on dynamic evolution particle swarm shielding EMD |
| CN113362853A (en) * | 2020-03-03 | 2021-09-07 | 东北大学秦皇岛分校 | EMD endpoint effect suppression method based on LSTM network |
| WO2022166401A1 (en) * | 2021-02-05 | 2022-08-11 | 中国科学院深圳先进技术研究院 | Eemd-pca-based method and device for removing motion artifact from eeg signal |
| CN115001517A (en) * | 2022-05-23 | 2022-09-02 | 四川轻化工大学 | Wireless communication receiving noise suppression method |
| CN115204243A (en) * | 2022-09-15 | 2022-10-18 | 西南交通大学 | LMD endpoint effect improvement method based on similar triangular waveform matching continuation |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103106903A (en) * | 2013-01-11 | 2013-05-15 | 太原科技大学 | Single channel blind source separation method |
| US20130294611A1 (en) * | 2012-05-04 | 2013-11-07 | Sony Computer Entertainment Inc. | Source separation by independent component analysis in conjuction with optimization of acoustic echo cancellation |
| CN104636314A (en) * | 2015-01-30 | 2015-05-20 | 太原科技大学 | Single channel chaotic signal blind source separation method |
-
2017
- 2017-12-14 CN CN201711393096.2A patent/CN108231087A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20130294611A1 (en) * | 2012-05-04 | 2013-11-07 | Sony Computer Entertainment Inc. | Source separation by independent component analysis in conjuction with optimization of acoustic echo cancellation |
| CN103106903A (en) * | 2013-01-11 | 2013-05-15 | 太原科技大学 | Single channel blind source separation method |
| CN104636314A (en) * | 2015-01-30 | 2015-05-20 | 太原科技大学 | Single channel chaotic signal blind source separation method |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108962276A (en) * | 2018-07-24 | 2018-12-07 | 北京三听科技有限公司 | A kind of speech separating method and device |
| CN108962276B (en) * | 2018-07-24 | 2020-11-17 | 杭州听测科技有限公司 | Voice separation method and device |
| CN111242366B (en) * | 2020-01-08 | 2023-10-31 | 广东技术师范大学 | EMD method and device capable of being used for processing signals in real time |
| CN111242366A (en) * | 2020-01-08 | 2020-06-05 | 广东技术师范大学 | EMD method and device for processing signals in real time |
| CN113314137A (en) * | 2020-02-27 | 2021-08-27 | 东北大学秦皇岛分校 | Mixed signal separation method based on dynamic evolution particle swarm shielding EMD |
| CN113314137B (en) * | 2020-02-27 | 2022-07-26 | 东北大学秦皇岛分校 | Mixed signal separation method based on dynamic evolution particle swarm shielding EMD |
| CN113362853A (en) * | 2020-03-03 | 2021-09-07 | 东北大学秦皇岛分校 | EMD endpoint effect suppression method based on LSTM network |
| CN111246285A (en) * | 2020-03-24 | 2020-06-05 | 北京奇艺世纪科技有限公司 | Method for separating sound in comment video and method and device for adjusting volume |
| CN111784068A (en) * | 2020-07-09 | 2020-10-16 | 北京理工大学 | EEMD-based power load combined prediction method and device |
| WO2022166401A1 (en) * | 2021-02-05 | 2022-08-11 | 中国科学院深圳先进技术研究院 | Eemd-pca-based method and device for removing motion artifact from eeg signal |
| CN115001517A (en) * | 2022-05-23 | 2022-09-02 | 四川轻化工大学 | Wireless communication receiving noise suppression method |
| CN115204243A (en) * | 2022-09-15 | 2022-10-18 | 西南交通大学 | LMD endpoint effect improvement method based on similar triangular waveform matching continuation |
| CN115204243B (en) * | 2022-09-15 | 2023-02-07 | 西南交通大学 | LMD endpoint effect improvement method based on similar triangular waveform matching continuation |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108231087A (en) | A kind of single channel blind source separating method | |
| CN103106903B (en) | Single channel blind source separation method | |
| Jutten et al. | Blind separation of sources, part I: An adaptive algorithm based on neuromimetic architecture | |
| O'grady et al. | Survey of sparse and non‐sparse methods in source separation | |
| Zibulevsky et al. | Blind source separation by sparse decomposition in a signal dictionary | |
| CN109890043B (en) | Wireless signal noise reduction method based on generative countermeasure network | |
| Pearlmutter et al. | A context-sensitive generalization of ICA | |
| CN110060657B (en) | SN-based many-to-many speaker conversion method | |
| CN104636314A (en) | Single channel chaotic signal blind source separation method | |
| Kuo et al. | Variational recurrent neural networks for speech separation | |
| Hansen et al. | On independent component analysis for multimedia signals | |
| CN113229829B (en) | Quaternion electroencephalogram signal extraction method and quaternion electroencephalogram signal extraction system | |
| Hamidi Ghalehjegh et al. | Fast block-sparse decomposition based on SL0 | |
| US7010514B2 (en) | Blind signal separation system and method, blind signal separation program and recording medium thereof | |
| Das et al. | ICA methods for blind source separation of instantaneous mixtures: A case study | |
| Missaoui et al. | Blind speech separation based on undecimated wavelet packet-perceptual filterbanks and independent component analysis | |
| CN110060699A (en) | A kind of single channel speech separating method based on the sparse expansion of depth | |
| CN103903631A (en) | Speech signal blind separating method based on variable step size natural gradient algorithm | |
| Peng et al. | Reducing permutation error in subband-based convolutive blind separation | |
| Houda et al. | A novel method based on gaussianity and sparsity for signal separation algorithms | |
| Szu et al. | Statistical mechanics demixing approach to selection of independent wavelet basis | |
| Obradovic et al. | An information theory based learning paradigm for linear feature extraction | |
| CN116540189A (en) | Radar main lobe forwarding interference resisting method based on reversible residual error network | |
| Zeinalzadeh et al. | A neural network model to classify liver cancer patients using data expansion and compression | |
| Margaris et al. | A parallel implementation of the natural gradient BSS method using MPI |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20180629 |
|
| WD01 | Invention patent application deemed withdrawn after publication |