CN108520756A - A kind of method and device of speaker's speech Separation - Google Patents
A kind of method and device of speaker's speech Separation Download PDFInfo
- Publication number
- CN108520756A CN108520756A CN201810231676.XA CN201810231676A CN108520756A CN 108520756 A CN108520756 A CN 108520756A CN 201810231676 A CN201810231676 A CN 201810231676A CN 108520756 A CN108520756 A CN 108520756A
- Authority
- CN
- China
- Prior art keywords
- audio signal
- audio
- obtains
- speaker
- result
- 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
- G10L21/0272—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
- 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
- G10L21/0272—Voice signal separating
- G10L21/028—Voice signal separating using properties of sound source
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
- G10L21/00—Processing of the speech or voice signal to produce another audible or non-audible signal, e.g. visual or tactile, in order to modify its quality or its intelligibility
- G10L21/02—Speech enhancement, e.g. noise reduction or echo cancellation
- G10L21/0208—Noise filtering
- G10L21/0216—Noise filtering characterised by the method used for estimating noise
- G10L2021/02161—Number of inputs available containing the signal or the noise to be suppressed
- G10L2021/02166—Microphone arrays; Beamforming
Abstract
The invention discloses a kind of method and device of speaker's speech Separation, method includes：Obtain the audio signal of preset format；It is pre-processed for the audio signal, first audio signal that obtains that treated；Audio separating treatment is carried out for first audio signal, obtains the second audio signal of different direction speaker；Enhancing processing is carried out for second audio signal, obtains the third audio signal of enhanced different direction speaker；Export the third audio signal.Technical solution using the present invention realizes quickly and accurately separation without the audio signal of multiple speakers in orientation.
Description
Technical field
The present invention relates to technical field of voice recognition, and in particular to a kind of method and device of speaker's speech Separation.
Background technology
With the development of science and technology, every field is higher and higher for the pursuit of audio quality, all kinds of audio documents
Acquiring way is more and more abundant, and data volume is in explosive growth, to also more and more difficult to the management of audio documents.In recent years
Come, people begin one's study audio retrieval technology, to the multi-media voices such as call voice, broadcasting speech and conference voice document into
Row management.Wherein, maximum to the retrieval difficulty of conference voice, because including multiple channels in conference voice document, more
Speaker.
Existing audio separation method is broadly divided into single channel (Mike) technology and multichannel (Mike) technology.Single Mike's skill
Art includes mainly the audio separation method based on model and the separation method based on distance scale；More Mike's technologies include mainly wave
Beam forms separation method and blind source separation method.
Wherein, the audio separation method based on model includes two steps of training and identification：To inputting sound in training process
Frequency carries out the laggard step of feature extraction and is trained and stores the model after training；Feature is carried out to input audio in identification process
After carrying out speaker's separation and speaker clustering after extraction, matching primitives are further carried out with the model of storage, judgement is each
Speaker finally obtains the audio signal after separation.Separation method based on distance scale then passes through the left and right of calculating every bit
The distance of two segment signals of adjacent certain window length, is further compared with the threshold value of setting, obtains the jump of audio signal
Height, to the audio signal after being detached.Wave beam forming separation method by input audio carry out auditory localization in real time,
And enhancing processing is further carried out according to speaker orientation, obtain the audio signal of each speaker.Blind source separation method passes through
Blind source separating processing is carried out to input audio, to obtain the audio signal of each speaker.
But the separation method based on model, it is desirable that the time that each speaker continuously speaks in dialogue is longer, and
Algorithm complexity is excessively high；The problems such as separation method based on distance scale, there are testing number excessive redundancy cut-points.And wave beam
Form separation method, the methods of blind source separation method, primarily directed to linear microphone array and plane microphone array etc. into
Row processing, and there are certain deficiencies for the effect handled in complex environment.
Therefore, under complex environment, more quickly and accurately separation does not have to the audio signal of multiple speakers in orientation,
It is current technical problem urgently to be resolved hurrily.
Invention content
The purpose of the present invention is to provide a kind of method and devices of speaker's speech Separation, realize quickly and accurately
Separation does not have to the audio signal of multiple speakers in orientation.
To achieve the above object, the present invention provides a kind of method of speaker's speech Separation, including：
Obtain the audio signal of preset format；
It is pre-processed for the audio signal, first audio signal that obtains that treated；
Audio separating treatment is carried out for first audio signal, obtains the second audio letter of different direction speaker
Number；
Enhancing processing is carried out for second audio signal, obtains the third audio of enhanced different direction speaker
Signal；
Export the third audio signal.
Further, it in method described above, is pre-processed for the audio signal, obtains that treated first
Audio signal, including：
Obtain the modes of emplacement parameter and ambient parameters of wheat battle array；
According to the modes of emplacement parameter of wheat battle array, conversion process is carried out to the audio signal, obtains being located at same flat
The transducing audio signal in face；
Time-frequency conversion is carried out to the transducing audio signal, obtains the corresponding frequency-region signal of the transducing audio signal；
According to the ambient parameters, audio enhancing processing is carried out to the frequency-region signal, obtains enhanced frequency domain
Signal；
Time-frequency inverse transformation is carried out for enhanced frequency-region signal, time-domain signal is obtained, as first audio signal.
Further, in method described above, audio separating treatment is carried out to first audio signal, obtains difference
The second audio signal of orientation speaker, including：
According to first audio signal, obtains the corresponding auditory localization result of first audio signal and speaker knows
Other result；
According to the auditory localization result and the Speaker Identification as a result, carrying out audio point to first audio signal
From processing, second audio signal is obtained.
Further, in method described above, according to first audio signal, first audio signal pair is obtained
The auditory localization result and Speaker Identification answered are as a result, include：
Speech detection processing is carried out to first audio signal, obtains testing result；
According to the testing result, auditory localization processing is carried out to first audio signal, obtains the auditory localization
As a result；
According to preset identification model, Speaker Identification processing is carried out to first audio signal, obtains described speak
People's recognition result.
Further, in method described above, according to the auditory localization result and the Speaker Identification as a result, right
First audio signal carries out audio separating treatment, obtains second audio signal, including：
According to the auditory localization result and the Speaker Identification as a result, using Beamforming Method, to described first
Audio signal carries out audio separating treatment, obtains second audio signal.
Further, in method described above, according to the auditory localization result and the Speaker Identification as a result, right
First audio signal carries out audio separating treatment, obtains second audio signal, including：
Choose audio separation method corresponding with the auditory localization result；
According to the Speaker Identification as a result, using the audio separation method, sound is carried out to first audio signal
Frequency separating treatment obtains second audio signal.
Further, in method described above, enhancing processing is carried out to second audio signal, is obtained enhanced
Third audio signal, including：
Based on the Speaker Identification as a result, second audio signal is smoothed and audio conversion point
The correcting process set obtains the third audio signal.
The present invention also provides a kind of devices of speaker's speech Separation, including：
Acquisition module, the audio signal for obtaining preset format；
Preprocessing module is pre-processed for being directed to the audio signal, first audio signal that obtains that treated；
Audio separation module obtains different direction and speaks for carrying out audio separating treatment to first audio signal
The second audio signal of people；
Enhancing processing module obtains enhanced third audio for carrying out enhancing processing to second audio signal
Signal；
Output module, for exporting the third audio signal.
The method and device of speaker's speech Separation of the present invention, is located in advance by the audio signal to preset format
Reason, first audio signal that obtains that treated carry out audio separating treatment to the first audio signal, obtain different direction speaker
The second audio signal, enhancing processing is carried out to the second audio signal, obtains the third sound of enhanced different direction speaker
Frequency signal exports third audio signal, realizes quickly and accurately separation without the audio signal of multiple speakers in orientation.
Description of the drawings
Fig. 1 is the flow chart of the embodiment of the method for speaker's speech Separation of the present invention；
Fig. 2 is the wheat battle array placement schematic diagram of the present invention four tunnel audio signals of acquisition；
Fig. 3 is the structural schematic diagram of the device embodiment of speaker's speech Separation of the present invention.
Specific implementation mode
To make the object, technical solutions and advantages of the present invention clearer, below in conjunction with the specific embodiment of the invention and
The present embodiment technical solution is clearly and completely described in corresponding attached drawing.Obviously, described embodiment is only this implementation
Example a part of the embodiment, instead of all the embodiments.Based on the embodiments of the present invention, those of ordinary skill in the art are not having
There is the every other embodiment obtained under the premise of making creative work, belongs to the range of the present embodiment protection.
The (if present)s such as term " first ", " second " in specification and claims and above-mentioned attached drawing are to be used for area
Not similar part, without being used to describe specific sequence or precedence.It should be appreciated that the data used in this way are appropriate
In the case of can be interchanged, so that embodiments herein described herein can be to the sequence other than illustrating herein
Implement.
The following examples are used to illustrate the present invention, but are not intended to limit the scope of the present invention..
Embodiment 1
Fig. 1 is the flow chart of the embodiment of the method for speaker's speech Separation of the present invention, as shown in Figure 1, the present embodiment
The method of speaker's speech Separation can specifically include following steps：
100, the audio signal of preset format is obtained.
The audio signal of preset format in the present embodiment can be the audio signal of Ambisonic A formats.Its
In, the audio signal of Ambisonic A formats is four tunnel audio signals (left front road (Left-Front-Up, LFU), You Qianlu
(Right-Front-Down, RFD), left back road (Left-Back-Down, LBD), the right way of escape (Right-Back-Up, RBU)).
Fig. 2 is the wheat battle array placement schematic diagram of the present invention four tunnel audio signals of acquisition.
101, it is pre-processed for the audio signal of acquisition, first audio signal that obtains that treated.
During specific implementation at one, when obtaining the audio signal of preset format, the placement side of wheat battle array can be obtained
Formula parameter and ambient parameters, so as to according to the modes of emplacement parameter of wheat battle array, to the audio signal of the preset format of acquisition into
Row conversion process, the transducing audio signal being generally aligned in the same plane, and time-frequency conversion is carried out to transducing audio signal, turned
The corresponding frequency-region signal of audio signal is changed, and, according to ambient parameters, audio enhancing processing is carried out to frequency-region signal, is obtained
To enhanced frequency-region signal, time-frequency inverse transformation further is carried out to frequency-region signal, obtains time-domain signal, believed as the first audio
Number.
For example, after getting the modes of emplacement of Mai Zhen, can the modes of emplacement based on wheat battle array to audio signal according to public affairs
Formula (1) carry out rotation processing, so as to get audio signal be generally aligned in the same plane.
Wherein, A is transition matrix：
Wherein, θhFor angle of heading, θpFor pitch angle, θbFor inclination angle, f (θh,θp,θb) be and θh、θpAnd θbRelevant letter
Number.
After obtaining conversion signal, discrete fourier transform (Discrete Fourier may be used
Transformation), the methods of fast Fourier transform (Fast Fourier Transformation, FFT) is to conversion signal
Time-frequency conversion processing is carried out by road.By taking DFT as an example, time-frequency conversion processing can be carried out to conversion signal according to formula (2)：
Wherein, AN domain index value when n is, k are frequency domain index value, and L is audio frequency process frame length, LfFor the length of time-frequency conversion, j
For imaginary part unit, M is number of channels, and x (n) is audio time domain sample value, and X (k) is audio frequency coefficient.
After obtaining frequency-region signal, reverberation time (RT can be passed through according to 4 tunnel audio signal estimated noise energy spectrums60)
Parameter and through and reflectivity (Direct-to-Reverberant Energy Ratio, DRR) parameter Estimation reverberation energy
Spectrum, the noise energy spectrum and reverberation energy spectrum for being based further on estimation carry out audio enhancing processing by road, to obtained frequency
Domain signal carries out the processing such as denoising, dereverberation, so as to get frequency-region signal enhanced.
It, can be according to the modes of emplacement parameter and ambient parameters of wheat battle array, to the multichannel sound of reception in the present embodiment
Frequency signal is pre-processed, and influence of the environment to follow audio separating treatment is reduced.
102, audio separating treatment is carried out to the first audio signal, obtains the second audio signal of different direction speaker.
In the present embodiment, after obtaining the first audio signal, first audio can be obtained according to first audio signal
The corresponding auditory localization result of signal and Speaker Identification are as a result, and according to auditory localization result and Speaker Identification as a result, right
First audio signal carries out audio separating treatment, to obtain the second audio signal of different direction speaker.
During specific implementation at one, speech detection processing can be carried out to the first audio signal, be examined accordingly
Survey as a result, so as to according to the testing result, auditory localization processing carried out to the first audio signal, obtain auditory localization as a result, with
And according to preset identification model, Speaker Identification processing is carried out to the first audio signal, obtains Speaker Identification result.
For example, may be used multiple signal classification (Multiple Signal Classification, MUSIC) algorithm,
The methods of broad sense cross-correlation (Generalized Cross Correlation, GCC) realizes auditory localization, specific by taking GCC as an example
It can realize in the following way：
A) cross-correlation of each road audio is calculated separately according to formula (3)：
Wherein, K1To originate frequency point, K2To end frequency point.
B) it is smoothed based on voice detection results according to formula (4)：
Gsm(i, j)=Gsm(i,j)*fsm+(1-fsm)*G(i,j) (4)
Wherein, fsmFor smoothing factor：
Vad is speech detection handling result.
C) cross-correlation function after smooth is further processed, obtains auditory localization result.
In the present embodiment, can the mode based on model carry out Speaker Identification, obtain Speaker Identification as a result, such as Gauss
Mixed model (Gaussian Mixed Model, GMM), Hidden Markov Model (Hidden Markov Model, HMM) are deep
Spend neural network (Deep Neural Networks, DNN) etc..
After obtaining auditory localization result and Speaker Identification result, Wave beam forming mode may be used, to first via sound
Frequency signal carries out audio separating treatment, obtains the second audio signal of different direction speaker.
Audio separation method corresponding with auditory localization result can also be chosen, and according to Speaker Identification as a result, profit
With audio separation method, audio separating treatment is carried out to the first audio signal, obtains the second audio letter of different direction speaker
Number.
For example, audio separating treatment can be carried out using formula (5), the second audio letter of different direction speaker is obtained
Number.
Wherein, VdoaFor in the weighted factor of Sounnd source direction：
τ is time delay, and S is sound source number, VspeFor simple sund source when weighted factor.
As S ＞ 1,Beamforming Method may be used and obtain the audio signal of Sounnd source direction.When S≤
When 1, Vdoa=Vspe, for example be set as showing when (1,0,0,0) using the 1st tunnel audio as the audio signal after separation.
102, enhancing processing is carried out to the second audio signal of different direction speaker, obtains enhanced different direction and says
Talk about the third audio signal of people.
For example, can be based on Speaker Identification as a result, the second audio signal to different direction speaker is smoothly located
The correcting process of reason and audio position of conversion point, obtains the third audio signal of different direction speaker, to ensure audio
Continuity.
103, third audio signal is exported.
The executive agent of the method for speaker's speech Separation of the present embodiment can be the device of speaker's speech Separation, should
The device of speaker's speech Separation can specifically be integrated by software, such as the device of speaker's speech Separation specifically can be with
It is applied for one, the present invention is to this without being particularly limited to.
The method of speaker's speech Separation of the present embodiment, obtains the audio signal of preset format, by audio signal
It is pre-processed, first audio signal that obtains that treated carries out audio separating treatment to the first audio signal, obtains not Tongfang
The second audio signal of position speaker, carries out enhancing processing to the second audio signal, obtains enhanced different direction speaker
Third audio signal, export third audio signal, realize quickly and accurately separation without orientation multiple speakers sound
Frequency signal.
Embodiment 2
Fig. 3 is the structural schematic diagram of the device embodiment of speaker's speech Separation of the present invention, as shown in figure 3, this implementation
The device of speaker's speech Separation of example may include acquisition module 10, preprocessing module 11, audio separation module 12, at enhancing
Manage module 13 and output module 14.
Wherein, acquisition module 10, the audio signal for obtaining preset format.
The audio signal of preset format in the present embodiment can be the audio signal of Ambisonic A formats.Its
In, the audio signal of Ambisonic A formats is four tunnel audio signals (left front road (Left-Front-Up, LFU), You Qianlu
(Right-Front-Down, RFD), left back road (Left-Back-Down, LBD), the right way of escape (Right-Back-Up, RBU)).
Fig. 2 is the wheat battle array placement schematic diagram of the present invention four tunnel audio signals of acquisition
Preprocessing module 11 is received audio signal and is pre-processed for docking, first audio signal that obtains that treated.Tool
Body, preprocessing module 11 can obtain the modes of emplacement parameter and ambient parameters of wheat battle array；According to the modes of emplacement of wheat battle array
Parameter carries out conversion process, the transducing audio signal being generally aligned in the same plane to multi-channel audio signal；To conversion signal into
Row time-frequency conversion obtains the corresponding frequency-region signal of conversion signal；According to ambient parameters, audio enhancing is carried out to frequency-region signal
Processing, obtains enhanced frequency-region signal；Time-frequency inverse transformation is carried out to enhanced audio signal, obtains audio time domain signal,
As the first audio signal.
Audio separation module 12 obtains different direction speaker for carrying out audio separating treatment to the first audio signal
The second audio signal.Specifically, audio separation module 12 can obtain the first audio signal and correspond to according to the first audio signal
Auditory localization result and Speaker Identification as a result, for example, to the first audio signal carry out speech detection processing, obtain detection knot
Fruit；According to testing result, auditory localization processing is carried out to the first audio signal, obtains auditory localization result；According to preset knowledge
Other model carries out Speaker Identification processing to the first audio signal, obtains Speaker Identification result.
Audio separation module 12 can also according to auditory localization result and Speaker Identification as a result, to the first audio signal into
Row audio separating treatment obtains the second audio signal of different direction speaker.For example, according to auditory localization result and can say
People's recognition result is talked about, using beam-forming technology, audio separating treatment is carried out to the first audio signal, obtains speaking without orientation
The second audio signal of people.Or choose audio separation method corresponding with auditory localization result；Known according to speaker
Not as a result, using audio separation method, audio separating treatment is carried out to the first audio signal, obtains the of different direction speaker
Two audio signals.
Enhance processing module 13, carries out enhancing processing for the second audio signal to different direction speaker, increased
The third audio signal of different direction speaker after strong.Specifically, enhancing processing module 13 can be based on Speaker Identification knot
Fruit, is smoothed the second audio signal and the correcting process of audio position of conversion point, obtains different direction speaker
Third audio signal.
Output module 14, the third audio signal for exporting different direction speaker.
The device of speaker's speech Separation of the present embodiment, by using the realization machine of above-mentioned each module separating audio signals
System is identical as the realization mechanism of above-mentioned embodiment illustrated in fig. 1, can refer to the record of above-mentioned embodiment illustrated in fig. 1 in detail, herein
It repeats no more.
The device of speaker's speech Separation of the present embodiment, obtains the audio signal of preset format, by audio signal
It is pre-processed, first audio signal that obtains that treated carries out audio separating treatment to the first audio signal, obtains not Tongfang
The second audio signal of position speaker, carries out enhancing processing to the second audio signal, obtains enhanced different direction speaker
Third audio signal, export third audio signal, realize quickly and accurately separation without orientation multiple speakers sound
Frequency signal.
Although above having used general explanation and specific embodiment, the present invention is described in detail, at this
On the basis of invention, it can be made some modifications or improvements, this will be apparent to those skilled in the art.Therefore,
These modifications or improvements without departing from theon the basis of the spirit of the present invention belong to the scope of protection of present invention.
Claims (8)
1. a kind of method of speaker's speech Separation, which is characterized in that including：
Obtain the audio signal of preset format；
It is pre-processed for the audio signal, first audio signal that obtains that treated；
Audio separating treatment is carried out for first audio signal, obtains the second audio signal of different direction speaker；
Enhancing processing is carried out for second audio signal, obtains the third audio letter of enhanced different direction speaker
Number；
Export the third audio signal.
2. according to the method described in claim 1, it is characterized in that, pre-processed for the audio signal, handled
The first audio signal afterwards, including：
Obtain the modes of emplacement parameter and ambient parameters of wheat battle array；
According to the modes of emplacement parameter of wheat battle array, conversion process is carried out to the audio signal, is generally aligned in the same plane
Transducing audio signal；
Time-frequency conversion is carried out to the transducing audio signal, obtains the corresponding frequency-region signal of the transducing audio signal；
According to the ambient parameters, audio enhancing processing is carried out to the frequency-region signal, obtains enhanced frequency-region signal；
Time-frequency inverse transformation is carried out for enhanced frequency-region signal, time-domain signal is obtained, as first audio signal.
3. method according to claim 1 or 2, which is characterized in that carried out at audio separation to first audio signal
Reason, obtains the second audio signal of different direction speaker, including：
According to first audio signal, the corresponding auditory localization result of first audio signal and Speaker Identification knot are obtained
Fruit；
According to the auditory localization result and the Speaker Identification as a result, being carried out at audio separation to first audio signal
Reason, obtains second audio signal.
4. according to the method described in claim 3, it is characterized in that, according to first audio signal, first sound is obtained
The corresponding auditory localization result of frequency signal and Speaker Identification are as a result, include：
Speech detection processing is carried out to first audio signal, obtains testing result；
According to the testing result, auditory localization processing is carried out to first audio signal, obtains the auditory localization result；
According to preset identification model, Speaker Identification processing is carried out to first audio signal, the speaker is obtained and knows
Other result.
5. according to the method described in claim 3, it is characterized in that, according to the auditory localization result and the Speaker Identification
As a result, carrying out audio separating treatment to first audio signal, second audio signal is obtained, including：
According to the auditory localization result and the Speaker Identification as a result, using Beamforming Method, to first audio
Signal carries out audio separating treatment, obtains second audio signal.
6. according to the method described in claim 3, it is characterized in that, according to the auditory localization result and the Speaker Identification
As a result, carrying out audio separating treatment to first audio signal, second audio signal is obtained, including：
Choose audio separation method corresponding with the auditory localization result；
According to the Speaker Identification as a result, using the audio separation method, audio point is carried out to first audio signal
From processing, second audio signal is obtained.
7. according to the method described in claim 3, it is characterized in that, carry out enhancing processing to second audio signal, obtain
Enhanced third audio signal, including：
Based on the Speaker Identification as a result, being smoothed to second audio signal and audio position of conversion point
Correcting process obtains the third audio signal.
8. a kind of device of speaker's speech Separation, which is characterized in that including：
Acquisition module, the audio signal for obtaining preset format；
Preprocessing module is pre-processed for being directed to the audio signal, first audio signal that obtains that treated；
Audio separation module obtains different direction speaker for carrying out audio separating treatment for first audio signal
The second audio signal；
Enhance processing module, enhancing processing is carried out for being directed to second audio signal, obtains enhanced third audio letter
Number；
Output module, for exporting the third audio signal.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810231676.XA CN108520756B (en) | 2018-03-20 | 2018-03-20 | Method and device for separating speaker voice |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810231676.XA CN108520756B (en) | 2018-03-20 | 2018-03-20 | Method and device for separating speaker voice |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108520756A true CN108520756A (en) | 2018-09-11 |
| CN108520756B CN108520756B (en) | 2020-09-01 |
Family
ID=63433795
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810231676.XA Active CN108520756B (en) | 2018-03-20 | 2018-03-20 | Method and device for separating speaker voice |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108520756B (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110021302A (en) * | 2019-03-06 | 2019-07-16 | 厦门快商通信息咨询有限公司 | A kind of Intelligent office conference system and minutes method |
| CN110459239A (en) * | 2019-03-19 | 2019-11-15 | 深圳壹秘科技有限公司 | Role analysis method, apparatus and computer readable storage medium based on voice data |
| CN111899758A (en) * | 2020-09-07 | 2020-11-06 | 腾讯科技（深圳）有限公司 | Voice processing method, device, equipment and storage medium |
| CN112382306A (en) * | 2020-12-02 | 2021-02-19 | 苏州思必驰信息科技有限公司 | Method and device for separating speaker audio |
| CN112634935A (en) * | 2021-03-10 | 2021-04-09 | 北京世纪好未来教育科技有限公司 | Voice separation method and device, electronic equipment and readable storage medium |
Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP1818909A1 (en) * | 2004-12-03 | 2007-08-15 | HONDA MOTOR CO., Ltd. | Voice recognition system |
| CN101720558A (en) * | 2007-04-19 | 2010-06-02 | 埃波斯开发有限公司 | Voice and position localization |
| CN102831898A (en) * | 2012-08-31 | 2012-12-19 | 厦门大学 | Microphone array voice enhancement device with sound source direction tracking function and method thereof |
| CN103456312A (en) * | 2013-08-29 | 2013-12-18 | 太原理工大学 | Single channel voice blind separation method based on computational auditory scene analysis |
| CN103811020A (en) * | 2014-03-05 | 2014-05-21 | 东北大学 | Smart voice processing method |
| CN104049235A (en) * | 2014-06-23 | 2014-09-17 | 河北工业大学 | Microphone array in sound source orienting device |
| CN104936091A (en) * | 2015-05-14 | 2015-09-23 | 科大讯飞股份有限公司 | Intelligent interaction method and system based on circle microphone array |
| CN105120421A (en) * | 2015-08-21 | 2015-12-02 | 北京时代拓灵科技有限公司 | Method and apparatus of generating virtual surround sound |
| CN105355203A (en) * | 2015-11-03 | 2016-02-24 | 重庆码头联智科技有限公司 | Method for speech judgment by virtue of gravity sensor intelligent wearable equipment |
| CN105872940A (en) * | 2016-06-08 | 2016-08-17 | 北京时代拓灵科技有限公司 | Virtual reality sound field generating method and system |
| CN106098075A (en) * | 2016-08-08 | 2016-11-09 | 腾讯科技（深圳）有限公司 | Audio collection method and apparatus based on microphone array |
| CN106816156A (en) * | 2017-02-04 | 2017-06-09 | 北京时代拓灵科技有限公司 | A kind of enhanced method and device of audio quality |
-
2018
- 2018-03-20 CN CN201810231676.XA patent/CN108520756B/en active Active
Patent Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP1818909A1 (en) * | 2004-12-03 | 2007-08-15 | HONDA MOTOR CO., Ltd. | Voice recognition system |
| CN101720558A (en) * | 2007-04-19 | 2010-06-02 | 埃波斯开发有限公司 | Voice and position localization |
| CN102831898A (en) * | 2012-08-31 | 2012-12-19 | 厦门大学 | Microphone array voice enhancement device with sound source direction tracking function and method thereof |
| CN103456312A (en) * | 2013-08-29 | 2013-12-18 | 太原理工大学 | Single channel voice blind separation method based on computational auditory scene analysis |
| CN103811020A (en) * | 2014-03-05 | 2014-05-21 | 东北大学 | Smart voice processing method |
| CN104049235A (en) * | 2014-06-23 | 2014-09-17 | 河北工业大学 | Microphone array in sound source orienting device |
| CN104936091A (en) * | 2015-05-14 | 2015-09-23 | 科大讯飞股份有限公司 | Intelligent interaction method and system based on circle microphone array |
| CN105120421A (en) * | 2015-08-21 | 2015-12-02 | 北京时代拓灵科技有限公司 | Method and apparatus of generating virtual surround sound |
| CN105355203A (en) * | 2015-11-03 | 2016-02-24 | 重庆码头联智科技有限公司 | Method for speech judgment by virtue of gravity sensor intelligent wearable equipment |
| CN105872940A (en) * | 2016-06-08 | 2016-08-17 | 北京时代拓灵科技有限公司 | Virtual reality sound field generating method and system |
| CN106098075A (en) * | 2016-08-08 | 2016-11-09 | 腾讯科技（深圳）有限公司 | Audio collection method and apparatus based on microphone array |
| CN106816156A (en) * | 2017-02-04 | 2017-06-09 | 北京时代拓灵科技有限公司 | A kind of enhanced method and device of audio quality |
Non-Patent Citations (2)
| Title |
|---|
| 张雄伟，李轶南，时文华，胡永刚，陈栩杉: "非负组合模型及其在声源分离中的应用", 《JOURNAL OF DATA ACQUISITION AND PROCESSING》 * |
| 陈洁: "背景音乐自动分离系统设计与实现", 《现代电子技术》 * |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110021302A (en) * | 2019-03-06 | 2019-07-16 | 厦门快商通信息咨询有限公司 | A kind of Intelligent office conference system and minutes method |
| CN110459239A (en) * | 2019-03-19 | 2019-11-15 | 深圳壹秘科技有限公司 | Role analysis method, apparatus and computer readable storage medium based on voice data |
| CN111899758A (en) * | 2020-09-07 | 2020-11-06 | 腾讯科技（深圳）有限公司 | Voice processing method, device, equipment and storage medium |
| CN111899758B (en) * | 2020-09-07 | 2024-01-30 | 腾讯科技（深圳）有限公司 | Voice processing method, device, equipment and storage medium |
| CN112382306A (en) * | 2020-12-02 | 2021-02-19 | 苏州思必驰信息科技有限公司 | Method and device for separating speaker audio |
| CN112382306B (en) * | 2020-12-02 | 2022-05-10 | 思必驰科技股份有限公司 | Method and device for separating speaker audio |
| CN112634935A (en) * | 2021-03-10 | 2021-04-09 | 北京世纪好未来教育科技有限公司 | Voice separation method and device, electronic equipment and readable storage medium |
| CN112634935B (en) * | 2021-03-10 | 2021-06-11 | 北京世纪好未来教育科技有限公司 | Voice separation method and device, electronic equipment and readable storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108520756B (en) | 2020-09-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108520756A (en) | A kind of method and device of speaker's speech Separation | |
| Chen et al. | Continuous speech separation: Dataset and analysis | |
| Yoshioka et al. | Multi-microphone neural speech separation for far-field multi-talker speech recognition | |
| CN110120227B (en) | Voice separation method of deep stack residual error network | |
| CN110970053B (en) | Multichannel speaker-independent voice separation method based on deep clustering | |
| Kingsbury et al. | Recognizing reverberant speech with RASTA-PLP | |
| CN106782565A (en) | A kind of vocal print feature recognition methods and system | |
| CN102565759B (en) | Binaural sound source localization method based on sub-band signal to noise ratio estimation | |
| CN107346664A (en) | A kind of ears speech separating method based on critical band | |
| Huang et al. | Audio replay spoof attack detection using segment-based hybrid feature and densenet-LSTM network | |
| Cai et al. | Multi-Channel Training for End-to-End Speaker Recognition Under Reverberant and Noisy Environment. | |
| CN110858476A (en) | Sound collection method and device based on microphone array | |
| Sainath et al. | Reducing the Computational Complexity of Multimicrophone Acoustic Models with Integrated Feature Extraction. | |
| Venkatesan et al. | Binaural classification-based speech segregation and robust speaker recognition system | |
| CN107895582A (en) | Towards the speaker adaptation speech-emotion recognition method in multi-source information field | |
| Taherian et al. | Multi-channel conversational speaker separation via neural diarization | |
| Kamble et al. | Teager energy subband filtered features for near and far-field automatic speech recognition | |
| Huang et al. | Audio-replay Attacks Spoofing Detection for Automatic Speaker Verification System | |
| Martín-Doñas et al. | Multi-channel block-online source extraction based on utterance adaptation | |
| CN113345421B (en) | Multi-channel far-field target voice recognition method based on angle spectrum characteristics | |
| Gaffar et al. | A multi-frame blocking for signal segmentation in voice command recognition | |
| CN114189781A (en) | Noise reduction method and system for double-microphone neural network noise reduction earphone | |
| Yang et al. | A target speaker separation neural network with joint-training | |
| Melhem et al. | Improving Deep Attractor Network by BGRU and GMM for Speech Separation | |
| Yoshioka et al. | Picknet: Real-time channel selection for ad hoc microphone arrays |
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 |