CN110111806A - A kind of blind separating method of moving source signal aliasing - Google Patents
A kind of blind separating method of moving source signal aliasing Download PDFInfo
- Publication number
- CN110111806A CN110111806A CN201910234231.1A CN201910234231A CN110111806A CN 110111806 A CN110111806 A CN 110111806A CN 201910234231 A CN201910234231 A CN 201910234231A CN 110111806 A CN110111806 A CN 110111806A
- Authority
- CN
- China
- Prior art keywords
- aliasing
- source signal
- signal
- time
- algorithm
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
- G10L21/00—Processing of the speech or voice signal to produce another audible or non-audible signal, e.g. visual or tactile, in order to modify its quality or its intelligibility
- G10L21/02—Speech enhancement, e.g. noise reduction or echo cancellation
- G10L21/0208—Noise filtering
- G10L21/0216—Noise filtering characterised by the method used for estimating noise
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
- G10L21/00—Processing of the speech or voice signal to produce another audible or non-audible signal, e.g. visual or tactile, in order to modify its quality or its intelligibility
- G10L21/02—Speech enhancement, e.g. noise reduction or echo cancellation
- G10L21/0208—Noise filtering
- G10L21/0216—Noise filtering characterised by the method used for estimating noise
- G10L21/0232—Processing in the frequency domain
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
- G10L21/00—Processing of the speech or voice signal to produce another audible or non-audible signal, e.g. visual or tactile, in order to modify its quality or its intelligibility
- G10L21/02—Speech enhancement, e.g. noise reduction or echo cancellation
- G10L21/0272—Voice signal separating
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
- G10L25/00—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00
- G10L25/48—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 specially adapted for particular use
- G10L25/69—Speech or voice analysis techniques not restricted to a single one of groups G10L15/00 - G10L21/00 specially adapted for particular use for evaluating synthetic or decoded voice signals
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
- G10L21/00—Processing of the speech or voice signal to produce another audible or non-audible signal, e.g. visual or tactile, in order to modify its quality or its intelligibility
- G10L21/02—Speech enhancement, e.g. noise reduction or echo cancellation
- G10L21/0208—Noise filtering
- G10L21/0216—Noise filtering characterised by the method used for estimating noise
- G10L2021/02161—Number of inputs available containing the signal or the noise to be suppressed
Abstract
The invention discloses a kind of blind separating methods of moving source signal aliasing, due in real life, the aliasing signal of acquisition has certain changeability, its aliasing filter is time-varying, compared with aliasing filter constant when traditional, the separation process of source signal will become more difficult, and the present invention is directed to the aliasing situation of this time-varying, propose a kind of blind separation algorithm of robust；First with the position of time difference algorithm for estimating positioning source signal, gives stringent mathematical theory and derive；Then aliasing filter is reconstructed, updates model parameter using expectation-maximization algorithm, recycles Wiener Filter Method source signals；Finally, the validity of proposed algorithm is verified by emulation experiment, a kind of moving source signal aliasing blind separation algorithm (Full-rank algorithm) of robust is compared simultaneously, it was demonstrated that the mentioned algorithm of the present invention has better robustness to mobile source signal aliasing blind separation.
Description
Technical field
The present invention relates to blind signal separation technology fields, and in particular to a kind of blind separating method of moving source signal aliasing.
Background technique
In classical cocktail party problem, since the people in party is in moving condition, the aliasing sound of acquisition compares
It is complicated.How original signal is only restored from the aliasing signal received, while aliasing channel is unknown, this provenance again
Signal separator process is referred to as " blind separation ".In recent years, blind separation technology has obtained sufficient application in speech signal processing.
But most research work, which is confined to source signal, does not move, that is, when constant aliasing signal, for the mixed of time-varying
The research of folded signal is relatively fewer.
In the prior art, document N.Q.K.Duong, E.Vincent, Under-determined reverberant
audio source separation using a full-rank spatial covariance model,IEEE
Trans.Audio Speech Lang.Process.18 (7) (2010) 1830-1840. proposes a kind of Full-rank algorithm pair
Mobile source signal aliasing blind separation has certain robustness, but effect is general, main cause be when source signal is mobile,
It also changes along with aliasing filter, causes the blind separation process of aliasing signal more difficult.However, in real life
In, such aliasing signal is generally existing.Therefore, how to propose that more efficient algorithm removes the source signal of separate mobile, still
It is so the hot spot of blind separation area research.
Summary of the invention
The purpose of the present invention is to overcome the shortcomings of the existing technology with it is insufficient, blind point of a kind of moving source signal aliasing is provided
From method, this method has better robustness to mobile source signal aliasing blind separation.
The purpose of the invention is achieved by the following technical solution:
A kind of blind separating method of moving source signal aliasing, includes the following steps:
Step 1 carries out mathematical modeling to mobile sound source signals aliasing；Assuming that going to connect using J fixed microphone
Receive the sound source signals of I movement, the mathematic(al) representation of the aliasing signal of generation are as follows:
Wherein, xjIt (t) is aliasing signal (j=1 ..., J) that j-th of microphone receives, hijt(τ) is the sky of time-varying
Between impulse response, τ is time delay；
In order to carry out the separation of source signal on frequency domain, Short Time Fourier Transform is carried out to above-mentioned formula (1), then time-varying
Space impulse responds hijtThe aliasing property of (τ) is slowly, it can be considered that when one very short with the variation of time
Between it is constant when aliasing process is in window；So the aliasing model formation (1) of time-varying can be similar to the aliasing model on frequency domain such as
Shown in following formula (2):
Wherein, xfn=[xfn1,...,xfnJ]TIndicate the Short Time Fourier Transform of aliasing signal, sfn,iIndicate i-th of source letter
Number single channel Short Time Fourier Transform, hfn,i=[hfn1,...,hfnJ]TIt is the space impulse response on frequency domain；
Step 2, the estimation of time delay τ；One group of microphone m and n are given, they are in the position of cartesian coordinate system point
It Wei not m and n, i.e. m ∈ R3,n∈R3, k0Refer to unit on the middle position p to sound source signals direction of two microphones to
Amount, i.e. k0∈R3,||k0| |=1, d are the distance between two microphones；Define p=[0,0,0] simultaneouslyTIt is cartesian coordinate
The origin fastened, θ refer to the elevation angle, and -90 °≤θ≤90 °；
Using vector product, can obtain:
Wherein, | | n | | indicate the norm of vector n, τn(k0) generate when indicating voice transmission to microphone n and position p when
Between postpone, v is the aerial spread speed of sound, takes v=340m/s；
It can be obtained according to above formula:
Therefore, the time delay between microphone m and n are as follows:
Wherein, m=1 ..., J, n=1 ..., J；
Step 3, the reconstruct of aliasing filter；Delay τ (m, n) between every group of microphone based on step 2 estimation is right
Should be in the phase difference on frequency domain:
hmn,i=exp (- j ωfτ (m, n)),
Wherein, ωf=2 π (f-1) Fs/ N, FsIt is sample frequency, N is Short Time Fourier Transform window length；
Step 4, the separation of source signal；Firstly, the Non-negative Matrix Factorization for defining the power spectral density of source signal is as follows:
Wherein,It is positive real value amplitude spectrogram, wfkIndicate single Non-negative Matrix Factorization component k's
Amplitude spectrum, hknIndicate the component gain of each frame n；
Then, Non-negative Matrix Factorization source signal model defined above is substituted into aliasing model formation (2), can be obtained:
Select square Frobenius norm as cost function, expression formula is as follows:
Then, model parameter H is updated using expectation-maximization algorithm, that is, EM algorithm iterationfn,wfk,hkn；Detail is such as
Under:
E-step: the conditional expectation counted naturally:
Rc,fn=diag ([wfkhkn]k),
M-step: the formula of model parameter is updated:
Wherein,And cfn=
[c1,fn,…,cK,fn]T∈CKThe vector of expression composition component coefficient, each component ck,fnPolynary multiple Gauss distribution appropriate is followed,
That is, ck,fn~Nc(0,wfkhkn)；
Meanwhile using Wiener Filter Method from aliasing signal source signals, obtain the estimation source signal on frequency domain:
Finally, the source signal of estimation is reconstructed in the inverse operation using Fourier transformation, the source signal in time domain is obtained.
The present invention have compared with prior art it is below the utility model has the advantages that
The present invention proposes a kind of blind separating method of moving source signal aliasing, first with time difference algorithm for estimating locating source
The position of signal gives stringent mathematical theory and derives；Then aliasing filter is reconstructed, is calculated using expectation maximization
Method updates model parameter, recycles Wiener Filter Method source signals；Finally, verifying the effective of mentioned algorithm by emulation experiment
Property, while comparing a kind of moving source signal aliasing blind separation algorithm (Full-rank algorithm) of robust, it was demonstrated that the present invention proposes calculation
Method has better robustness to mobile source signal aliasing blind separation；In addition, the present invention can position source signal well
Theory support has been established for the reconstruction of aliasing filter in position, efficiently avoids sequence ambiguity problem, improves source signal
Separating property, especially the separating effect of source signal is significant under low reverberation time environment, when simultaneously for higher echoing
Between environment have certain robustness.
Detailed description of the invention
Fig. 1 is flow chart of the invention；
Fig. 2 is that source signal of the invention positions schematic diagram；
Fig. 3 is source signal motion track schematic diagram of the invention；
Fig. 4 is the STOI evaluation result schematic diagram of reverberation time 150ms of the present invention；
Fig. 5 is the fwSegSNR evaluation result schematic diagram of reverberation time 150ms of the present invention；
Fig. 6 is the STOI evaluation result schematic diagram of reverberation time 200ms of the present invention；
Fig. 7 is the fwSegSNR evaluation result structural schematic diagram of reverberation time 200ms of the present invention.
Specific embodiment
Present invention will now be described in further detail with reference to the embodiments and the accompanying drawings, but embodiments of the present invention are unlimited
In this.
The invention proposes a kind of blind separating methods of moving source signal aliasing, since in real life, acquisition mixes
Folded signal has certain changeability, for example, in cocktail party problem some people be it is mobile, lead to the aliasing signal of acquisition
It is as made of mobile sound source signals aliasing.Therefore, its aliasing filter is time-varying, with it is traditional when constant aliasing
Filter is compared, and the separation process of source signal will become more difficult, and the present invention is directed to the aliasing situation of this time-varying, proposes one
The blind separation algorithm of kind robust.Firstly, carrying out mathematical modeling to moving source signal aliasing problem, ideal mathematical model is provided；
Then, the position that difference algorithm positioning source signal is reached based on the time is gone estimation time delay, recycles the time delay weight of estimation
Build aliasing filter；Finally, utilizing the real-time source signals of Wiener Filter Method.
As shown in Figure 1, a kind of blind separating method of moving source signal aliasing, includes the following steps:
Step 1 carries out mathematical modeling to mobile sound source signals aliasing；Assuming that going to connect using J fixed microphone
Receive the sound source signals of I movement, the mathematic(al) representation of the aliasing signal of generation are as follows:
Wherein, xjIt (t) is aliasing signal (j=1 ..., J) that j-th of microphone receives, hijt(τ) is the sky of time-varying
Between impulse response, τ is time delay；The purpose of the present invention is in aliasing filter hijtIn the case that (τ) is unknown, according only to connecing
The aliasing signal x receivedj(t), j=1 ..., J removes estimation source signal si(t), i=1 ..., I.
In order to carry out the separation of source signal on frequency domain, Short Time Fourier Transform is carried out to above-mentioned formula (1), then time-varying
Space impulse responds hijtThe aliasing property of (τ) is slowly, it can be considered that when one very short with the variation of time
Between it is constant when aliasing process is in window；So the aliasing model formation (1) of time-varying can be similar to the aliasing model on frequency domain such as
Shown in following formula (2):
Wherein, xfn=[xfn1,...,xfnJ]TIndicate the Short Time Fourier Transform of aliasing signal, sfn,iIndicate i-th of source letter
Number single channel Short Time Fourier Transform, hfn,i=[hfn1,...,hfnJ]TIt is the space impulse response on frequency domain；
Step 2, the estimation of time delay τ；As shown in Fig. 2, giving one group of microphone m and n, they are in cartesian coordinate
The position of system is respectively m and n, i.e. m ∈ R3,n∈R3, k0On the middle position p to sound source signals direction for referring to two microphones
Unit vector, i.e. k0∈R3,||k0| |=1, d are the distance between two microphones；Define p=[0,0,0] simultaneouslyTIt is flute
Origin on karr coordinate system, θ refer to the elevation angle, and -90 °≤θ≤90 °；
As shown in Fig. 2, can be obtained using vector product:
Wherein, | | n | | indicate the norm of vector n, τn(k0) generate when indicating voice transmission to microphone n and position p when
Between postpone, v is the aerial spread speed of sound, takes v=340m/s；
It can be obtained according to above formula:
Therefore, the time delay between microphone m and n are as follows:
Wherein, m=1 ..., J, n=1 ..., J；
Step 3, the reconstruct of aliasing filter；Delay τ (m, n) between every group of microphone based on step 2 estimation is right
Should be in the phase difference on frequency domain:
hmn,i=exp (- j ωfτ (m, n)),
Wherein, ωf=2 π (f-1) Fs/ N, FsIt is sample frequency, N is Short Time Fourier Transform window length；
Step 4, the separation of source signal；Firstly, the Non-negative Matrix Factorization for defining the power spectral density of source signal is as follows:
Wherein,It is positive real value amplitude spectrogram, wfkIndicate single Non-negative Matrix Factorization component k's
Amplitude spectrum, hknIndicate the component gain of each frame n；
Then, Non-negative Matrix Factorization source signal model defined above is substituted into aliasing model formation (2), can be obtained:
Select square Frobenius norm as cost function, expression formula is as follows:
Then, model parameter H is updated using expectation-maximization algorithm, that is, EM algorithm iterationfn,wfk,hkn；Detail is such as
Under:
E-step: the conditional expectation counted naturally:
Rc,fn=diag ([wfkhkn]k),
M-step: the formula of model parameter is updated:
Wherein,And cfn=
[c1,fn,…,cK,fn]T∈CKThe vector of expression composition component coefficient, each component ck,fnPolynary multiple Gauss distribution appropriate is followed,
That is, ck,fn~Nc(0,wfkhkn)；
Meanwhile using Wiener Filter Method from aliasing signal source signals, obtain the estimation source signal on frequency domain:
Finally, the source signal of estimation is reconstructed in the inverse operation using Fourier transformation, the source signal in time domain is obtained.
Illustrate the feasibility and superiority of inventive algorithm, all emulation below by two groups of specific emulation embodiments
Experiment is in Ubuntu15.04, Inter (R) Xeon (R) CPU E5-2630v3@2.40GHz, 32.00GB, Matlab
Programming is realized under R2016b environment.
Embodiment one:
In example 1, consider the aliasing situation that three mobile source signals are received by two microphones, the shifting of source signal
For dynamic rail mark as shown in figure 3, the distance between two microphones are 5 centimetres, reverberation time is 150 milliseconds；In order to evaluate blind separation
The quality of performance selects target sharpness measurement (STOI) in short-term and frequency weighting segmental signal-to-noise ratio (fwSegSNR) as evaluation
Criterion, experimental result are as shown in Figure 4 and Figure 5, it can be seen that, STOI the and fwSegSNR value ratio Full- that mentioned algorithm obtains
The result that rank algorithm obtains is more preferable, illustrates that the mentioned algorithm of the present invention has better Shandong to the blind separation of moving source signal aliasing
Stick.
Embodiment two:
In example 2, it also is contemplated that the aliasing situation of three mobile source signals, source signal are received by two microphones
Mobile space as shown in figure 3, the distance between two microphones be 5 centimetres, select reverberation time for 200 milliseconds；Experiment knot
Fruit is as shown in Figure 6 and Figure 7, and from the point of view of obtained STOI value and fwSegSNR value, mentioned algorithm separating resulting still compares Full-
The result that rank algorithm obtains is more preferable；In addition, the result of comparative example one, it is found that since the increase of reverberation time is led
Cause the decline of separating resulting.Therefore, under relatively low reverberation time, the mentioned algorithm of the present invention is mixed for mobile source signal
Folded blind separation has better separating property.
The present invention proposes a kind of blind separating method of moving source signal aliasing, first with time difference algorithm for estimating locating source
The position of signal gives stringent mathematical theory and derives；Then aliasing filter is reconstructed, is calculated using expectation maximization
Method updates model parameter, recycles Wiener Filter Method source signals；Finally, verifying the effective of mentioned algorithm by emulation experiment
Property, while comparing a kind of moving source signal aliasing blind separation algorithm (Full-rank algorithm) of robust, it was demonstrated that the present invention proposes calculation
Method has better robustness to mobile source signal aliasing blind separation；In addition, the present invention can position source signal well
Theory support has been established for the reconstruction of aliasing filter in position, efficiently avoids sequence ambiguity problem, improves source signal
Separating property, especially the separating effect of source signal is significant under low reverberation time environment, when simultaneously for higher echoing
Between environment have certain robustness.
Above-mentioned is the preferable embodiment of the present invention, but embodiments of the present invention are not limited by the foregoing content,
His any changes, modifications, substitutions, combinations, simplifications made without departing from the spirit and principles of the present invention, should be
The substitute mode of effect, is included within the scope of the present invention.
Claims (1)
1. a kind of blind separating method of moving source signal aliasing, which is characterized in that include the following steps:
Step 1 carries out mathematical modeling to mobile sound source signals aliasing；Assuming that going to receive I using J fixed microphone
The sound source signals of a movement, the mathematic(al) representation of the aliasing signal of generation are as follows:
Wherein, xjIt (t) is aliasing signal (j=1 ..., J) that j-th of microphone receives, hijt(τ) is the space impulse of time-varying
Response, τ is time delay；
In order to carry out the separation of source signal on frequency domain, Short Time Fourier Transform is carried out to above-mentioned formula (1), then the space of time-varying
Impulse response hijtThe aliasing property of (τ) is slowly, it can be considered that in a very short time window with the variation of time
Interior aliasing process is constant when being；So can be similar to the aliasing model on frequency domain for example following for the aliasing model formation (1) of time-varying
Shown in formula (2):
Wherein, xfn=[xfn1,...,xfnJ]TIndicate the Short Time Fourier Transform of aliasing signal, sfn,iIndicate i-th of source signal
Single channel Short Time Fourier Transform, hfn,i=[hfn1,...,hfnJ]TIt is the space impulse response on frequency domain；
Step 2, the estimation of time delay τ；One group of microphone m and n are given, they are respectively m in the position of cartesian coordinate system
And n, i.e. m ∈ R3,n∈R3, k0Refer to the unit vector on the middle position p to sound source signals direction of two microphones, i.e. k0
∈R3,||k0| |=1, d are the distance between two microphones；Define p=[0,0,0] simultaneouslyTIt is the original that cartesian coordinate is fastened
Point, θ refer to the elevation angle, and -90 °≤θ≤90 °；
Using vector product, can obtain:
Wherein, | | n | | indicate the norm of vector n, τn(k0) time for generating when indicating voice transmission to microphone n and position p prolongs
Late, v is the aerial spread speed of sound, takes v=340m/s；
It can be obtained according to above formula:
Therefore, the time delay between microphone m and n are as follows:
Wherein, m=1 ..., J, n=1 ..., J；
Step 3, the reconstruct of aliasing filter；Delay τ (m, n) between every group of microphone based on step 2 estimation, corresponds to
Phase difference on frequency domain is:
hmn,i=exp (- j ωfτ (m, n)),
Wherein, ωf=2 π (f-1) Fs/ N, FsIt is sample frequency, N is Short Time Fourier Transform window length；
Step 4, the separation of source signal；Firstly, the Non-negative Matrix Factorization for defining the power spectral density of source signal is as follows:
Wherein,It is positive real value amplitude spectrogram, wfkIndicate the amplitude of single Non-negative Matrix Factorization component k
Spectrum, hknIndicate the component gain of each frame n；
Then, Non-negative Matrix Factorization source signal model defined above is substituted into aliasing model formation (2), can be obtained:
Select square Frobenius norm as cost function, expression formula is as follows:
Then, model parameter H is updated using expectation-maximization algorithm, that is, EM algorithm iterationfn,wfk,hkn；Detail is as follows:
E-step: the conditional expectation counted naturally:
Rc,fn=diag ([wfkhkn]k)
M-step: the formula of model parameter is updated:
Wherein,And cfn=
[c1,fn,…,cK,fn]T∈CKThe vector of expression composition component coefficient, each component ck,fnPolynary multiple Gauss distribution appropriate is followed,
That is, ck,fn~Nc(0,wfkhkn)；
Meanwhile using Wiener Filter Method from aliasing signal source signals, obtain the estimation source signal on frequency domain:
Finally, the source signal of estimation is reconstructed in the inverse operation using Fourier transformation, the source signal in time domain is obtained.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910234231.1A CN110111806B (en) | 2019-03-26 | 2019-03-26 | Blind separation method for aliasing of mobile source signals |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910234231.1A CN110111806B (en) | 2019-03-26 | 2019-03-26 | Blind separation method for aliasing of mobile source signals |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110111806A true CN110111806A (en) | 2019-08-09 |
| CN110111806B CN110111806B (en) | 2021-03-30 |
Family
ID=67484552
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910234231.1A Active CN110111806B (en) | 2019-03-26 | 2019-03-26 | Blind separation method for aliasing of mobile source signals |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110111806B (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112599144A (en) * | 2020-12-03 | 2021-04-02 | Oppo（重庆）智能科技有限公司 | Audio data processing method, audio data processing apparatus, medium, and electronic device |
| WO2021226999A1 (en) * | 2020-05-15 | 2021-11-18 | Harman International Industries, Incorporated | Efficient blind source separation using topological approach |
| CN114220453A (en) * | 2022-01-12 | 2022-03-22 | 中国科学院声学研究所 | Multi-channel non-negative matrix decomposition method and system based on frequency domain convolution transfer function |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106504763A (en) * | 2015-12-22 | 2017-03-15 | 电子科技大学 | Based on blind source separating and the microphone array multiple target sound enhancement method of spectrum-subtraction |
| US20180190311A1 (en) * | 2015-06-30 | 2018-07-05 | Nec Corporation | Signal processing apparatus, signal processing method, and signal processing program |
| CN108269583A (en) * | 2017-01-03 | 2018-07-10 | 中国科学院声学研究所 | A kind of speech separating method based on time delay histogram |
-
2019
- 2019-03-26 CN CN201910234231.1A patent/CN110111806B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180190311A1 (en) * | 2015-06-30 | 2018-07-05 | Nec Corporation | Signal processing apparatus, signal processing method, and signal processing program |
| CN106504763A (en) * | 2015-12-22 | 2017-03-15 | 电子科技大学 | Based on blind source separating and the microphone array multiple target sound enhancement method of spectrum-subtraction |
| CN108269583A (en) * | 2017-01-03 | 2018-07-10 | 中国科学院声学研究所 | A kind of speech separating method based on time delay histogram |
Non-Patent Citations (3)
| Title |
|---|
| SHENGLI XIE, LIU YANG, JUN-MEI YANG, GUOXU ZHOU, AND YONG XIANG: "Time-frequency approach to un-derdetermined blind source separation", 《IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS》 * |
| YUAN XIE,KAN XIE,JUNJIE YANG AND SHENGLI XIE: "Underdetermined Blind Source Separation", 《SUMMER Y-BASEL》 * |
| YUAN XIE，KAN XIE，JUNJIE YANG，ZONGZE WU，SHENGLI XIE: "Underdetermined Reverberant Audio-Source Separation", 《CIRCUITS, SYSTEMS, AND SIGNAL PROCESSING》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2021226999A1 (en) * | 2020-05-15 | 2021-11-18 | Harman International Industries, Incorporated | Efficient blind source separation using topological approach |
| CN112599144A (en) * | 2020-12-03 | 2021-04-02 | Oppo（重庆）智能科技有限公司 | Audio data processing method, audio data processing apparatus, medium, and electronic device |
| CN114220453A (en) * | 2022-01-12 | 2022-03-22 | 中国科学院声学研究所 | Multi-channel non-negative matrix decomposition method and system based on frequency domain convolution transfer function |
| CN114220453B (en) * | 2022-01-12 | 2022-08-16 | 中国科学院声学研究所 | Multi-channel non-negative matrix decomposition method and system based on frequency domain convolution transfer function |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110111806B (en) | 2021-03-30 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110111806A (en) | A kind of blind separating method of moving source signal aliasing | |
| Haardt et al. | Simultaneous Schur decomposition of several nonsymmetric matrices to achieve automatic pairing in multidimensional harmonic retrieval problems | |
| CN109830245A (en) | A kind of more speaker's speech separating methods and system based on beam forming | |
| CN103854660B (en) | A kind of four Mike's sound enhancement methods based on independent component analysis | |
| EP3633676A1 (en) | Rnn-based noise reduction method and device for real-time conference | |
| CN105301563B (en) | A kind of double sound source localization method that least square method is converted based on consistent focusing | |
| CN101667425A (en) | Method for carrying out blind source separation on convolutionary aliasing voice signals | |
| CN112615801B (en) | Channel estimation method, medium, and apparatus based on compressed sensing and deep learning | |
| CN106961403B (en) | OFDM underwater acoustic voice communication Doppler compensation estimation method based on sparse channel model | |
| Wu et al. | Joint pitch and DOA estimation using the ESPRIT method | |
| CN101685638A (en) | Method and device for enhancing voice signals | |
| CN105929386A (en) | Wave arrival estimation method based on high-order accumulated amount | |
| CN109859749A (en) | A kind of voice signal recognition methods and device | |
| CN109166590B (en) | Two-dimensional time-frequency mask estimation modeling method based on spatial correlation | |
| CN109658948A (en) | One kind is towards the movable acoustic monitoring method of migratory bird moving | |
| CN109884592A (en) | A kind of auditory localization emulation mode towards low frequency Gaussian noise source | |
| CN110444220B (en) | Multi-mode remote voice perception method and device | |
| CN110111802A (en) | Adaptive dereverberation method based on Kalman filtering | |
| Zhong et al. | Direction of arrival tracking of an underwater acoustic source using particle filtering: Real data experiments | |
| CN116559778B (en) | Vehicle whistle positioning method and system based on deep learning | |
| CN110838303B (en) | Voice sound source positioning method using microphone array | |
| Tourbabin et al. | Speaker localization by humanoid robots in reverberant environments | |
| CN109658944B (en) | Helicopter acoustic signal enhancement method and device | |
| CN116992263A (en) | Method for extracting, identifying and separating self-noise characteristics of underwater vehicle based on multitask learning | |
| CN109541572B (en) | Subspace orientation estimation method based on linear environment noise model |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |