CN109859768A - Artificial cochlea's sound enhancement method - Google Patents
Artificial cochlea's sound enhancement method Download PDFInfo
- Publication number
- CN109859768A CN109859768A CN201910184264.XA CN201910184264A CN109859768A CN 109859768 A CN109859768 A CN 109859768A CN 201910184264 A CN201910184264 A CN 201910184264A CN 109859768 A CN109859768 A CN 109859768A
- Authority
- CN
- China
- Prior art keywords
- voice signal
- noise
- artificial cochlea
- sound
- scene
- 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.)
- Withdrawn
Links
Landscapes
- Soundproofing, Sound Blocking, And Sound Damping (AREA)
Abstract
The invention discloses a kind of artificial cochlea's sound enhancement methods comprising following steps: (A) pre-processes voice signal;(B) voice signal is subjected to end-point detection, judges noise frame and speech frame in voice signal, and the two is separated;(C) noise frame is subjected to feature extraction, extracts feature vector;(D) this feature vector is subjected to CNN operation, identifies noise scene;(E) corresponding gain table G (γ is selectedk,ξk), and speech enhan-cement parameter is chosen by calculating prior weight and posteriori SNR, voice signal is then subjected to speech enhan-cement;(F) encoded information output is converted by voice signal.Artificial cochlea's sound enhancement method gain table different by training, different speech enhan-cement parameters is made it have to adapt to different auditory scenes, the stimulus signal being more consistent with practical auditory scene is exported, clarity, the intelligibility of the voice signal of patient in a noisy environment are improved.
Description
Technical field
The present invention relates to a kind of sound enhancement method more particularly to a kind of artificial cochlea's sound enhancement methods.
Background technique
Artificial cochlea is recognized in the world bilateral severe or pole profound sensorineural hearing loss patient to be made to restore to listen
The unique effective ways and device felt.Existing artificial cochlea's operation workflow are as follows: sound is first converted to telecommunications by microphone acquisition
Number, it by special digitized processing, is encoded according still further to certain strategy, is transmitted to body by being loaded in the transmitting coil after ear
It is interior, it after the receiving coil of implant senses signal, is decoded by decoding chip, the stimulating electrode of implant is made to generate electric current,
To stimulate auditory nerve to generate the sense of hearing.Due to the limitation of use environment, environment noise is necessarily adulterated in sound, is needed to sound
Signal carries out certain enhancing optimization (i.e. noise reduction optimization), but in view of the diversification of use environment, traditional enhancing optimization does not have
There is universality, the signal after traditional enhancing optimization is deviated with actual conditions sometimes, is unable to reach optimal sense of hearing effect
Fruit.
Summary of the invention
In view of the above drawbacks of the prior art, technical problem to be solved by the invention is to provide a kind of artificial cochlea's languages
Sound Enhancement Method, with different speech enhan-cement parameters to adapt to different auditory scenes.
To achieve the above object, the present invention provides a kind of artificial cochlea's sound enhancement methods comprising following steps:
(A) preprocessor module pre-processes voice signal;(B) end-point detection program module is by pretreated voice signal
End-point detection is carried out, judges the noise frame and speech frame in voice signal, and the two is separated;(C) feature extraction program mould
The noise frame is carried out feature extraction by block, extracts feature vector;(D) this feature vector is carried out CNN by scene Recognition program module
Operation obtains the probability value of each default scene, and the maximum scene of probability value is determined as noise scene;(E) noise reduction process program
Module selects corresponding gain table G (γ according to the noise scenek,ξk), and by calculating prior weight and posteriori SNR choosing
Take gain table G (γk,ξk) in concrete sound enhance parameter, then by pretreated voice signal carry out speech enhan-cement;
(F) enhanced voice signal is converted encoded information output by tactful coded program module.
In step, which includes framing, adding window, preemphasis and its frequency domain conversion, wherein the adding window uses
Hamming window or Hanning window.
In stepb, which uses double threshold method, wherein the double threshold method is energy threshold and zero-crossing rate door
Limit.
In step C, this feature, which is extracted, uses MFCC, FBank or sound spectrograph.
In step E, gain table G (γk,ξk) γk, ξkRange take -19dB-20dB, training step is as follows:
(a) local minimum of signals with noise power spectrum is tracked:
(b) voice existing probability: Sr=P (λ, k)/P is calculatedmin(λ, k),
δ (k) is to determine empirical value related with frequency by experiment, is judged as that voice has frequency if Sr > δ (k)
Otherwise band is grass, it may be assumed that
According to the judgment rule of above formula, carry out more new speech existing probability:
P (λ, k)=αpp(λ-1,k)+(1-αp) I (λ, k), wherein αp=0.2;
(c) smoothing constant related with frequency: α is calculateds(λ, k)=αd+(1-αd) p (λ, k), wherein αd=0.85, αs
The value range of (λ, k) is αd≤αs(λ,k)≤1;
(d) noise power spectrum: D (λ, k)=α is updateds(λ,k)D(λ-1,k)+(1-αs(λ, k)) Shu Y (λ, k) Shu2, wherein D
(λ, k) is the estimated value of noise power spectrum, and wherein gain, which calculates, uses the least-mean-square error algorithm based on logarithmic spectrum
(LOGSTAS-MMSE), the amplitude spectrum estimator of voice:
Gain function:
Above formula integration type can approximate calculation, i.e.,
Further, the calculating formula of the prior weight isThe calculating formula of the posteriori SNR isWherein, λs(k)、λnIt (k) is the variance of voice and noise under k-th of frequency band, Y is noisy speech.
Artificial cochlea's sound enhancement method of the present invention gain table different by training, makes it have different speech enhan-cements
Parameter exports the stimulus signal being more consistent with practical auditory scene to adapt to different auditory scenes, improves patient in noise
Clarity, the intelligibility of voice signal under environment, improve the quality of life of artificial cave patient.
It is described further below with reference to technical effect of the attached drawing to design of the invention, specific structure and generation, with
It is fully understood from the purpose of the present invention, feature and effect.
Detailed description of the invention
Fig. 1 is the flow diagram of artificial cochlea's sound enhancement method of the present invention.
Specific embodiment
The present invention provides a kind of artificial cochlea's sound enhancement method, which can judge to be presently in ring
Border, and according to the corresponding speech enhan-cement parameter of the environmental selection judged, so that cochlear implant is obtained preferable sense of hearing sense
By.
As shown in Figure 1, artificial cochlea's sound enhancement method includes pretreatment, end-point detection, feature extraction, scene knowledge
Not, noise reduction process, strategy six steps of coding.
Pretreatment: voice signal is carried out framing, adding window, preemphasis and its frequency domain and converted by preprocessor module, wherein
The adding window uses Hamming window or Hanning window, and the purpose of frequency domain conversion is to carry out the conversion of time domain to frequency domain.
End-point detection: pretreated voice signal is carried out endpoint inspection using double threshold method by end-point detection program module
Survey, judge the noise frame and speech frame in voice signal, and the two is separated, wherein the double threshold method be energy threshold and
Zero-crossing rate thresholding.
Feature extraction: the noise frame is carried out feature extraction by feature extraction program module, is extracted feature vector, is used
MFCC (Mel-Frequency Cepstrum Coefficient, mel-frequency cepstrum coefficient), FBank (Mel-scale
Filter Bank, Meier scale filter group) or sound spectrograph.
Scene Recognition: this feature vector is carried out CNN (ConvolutionNeural by scene Recognition program module
Network, convolutional neural networks) operation, it obtains the probability value of each default scene, the maximum scene of probability value is determined as noise
Scene.
Noise reduction process: noise reduction process program module selects corresponding gain table G (γ according to the noise scenek,ξk), and lead to
It crosses calculating prior weight and posteriori SNR chooses gain table G (γk,ξk) in concrete sound enhance parameter, then to pre-
Treated, and voice signal carries out speech enhan-cement, to achieve the purpose that noise reduction.
Gain table G (γk,ξk), γk, ξkRange take -19dB-20dB, preferably take 0.5dB, training step is such as
Under:
(1) local minimum of signals with noise power spectrum is tracked:
(2) voice existing probability: Sr=P (λ, k)/P is calculatedmin(λ, k),
δ (k) is to determine empirical value related with frequency by experiment, is judged as that voice has frequency if Sr > δ (k)
Otherwise band is grass, it may be assumed that
According to the judgment rule of above formula, carry out more new speech existing probability:
P (λ, k)=αpp(λ-1,k)+(1-αp) I (λ, k), wherein αp=0.2;
(3) smoothing constant related with frequency: α is calculateds(λ, k)=αd+(1-αd) p (λ, k), wherein αd=0.85, αs
The value range of (λ, k) is αd≤αs(λ,k)≤1;
(4) noise power spectrum: D (λ, k)=α is updateds(λ,k)D(λ-1,k)+(1-αs(λ, k)) Shu Y (λ, k) Shu2, wherein D
(λ, k) is the estimated value of noise power spectrum, and wherein gain, which calculates, uses the least-mean-square error algorithm based on logarithmic spectrum
(LOGSTAS-MMSE), the amplitude spectrum estimator of voice:
Gain function:
Above formula integration type can approximate calculation, i.e.,
The calculating formula of the prior weight isThe calculating formula of the posteriori SNR isIts
In, λs(k)、λnIt (k) is the variance of voice and noise under k-th of frequency band, Y is noisy speech.
Strategy coding: tactful coded program module converts encoded information for the voice signal enhanced and exports, step packet
Envelope extraction is included, frequency spectrum choosing is big, non-linear compression, coding output, so that encoded information is sent in vivo by transmitting coil
Implant.
It has to be noted that the sound enhancement method can identify auditory scene in real time, it, can be automatic when auditory scene variation
Using new speech enhan-cement parameter.And the model of the CNN and gain table G (γk,ξk) lower training is completed online, actually make
Small with hour operation quantity, algorithm accounts for that hardware resource is few, and algorithmic delay is low, can run in the high mobile device of portability.
The preferred embodiment of the present invention has been described in detail above.It should be appreciated that those skilled in the art without
It needs creative work according to the present invention can conceive and makes many modifications and variations.Therefore, all technologies in the art
Personnel are available by logical analysis, reasoning, or a limited experiment on the basis of existing technology under this invention's idea
Technical solution, all should be within the scope of protection determined by the claims.
Claims (6)
1. a kind of artificial cochlea's sound enhancement method comprising following steps: (A) preprocessor module carries out voice signal
Pretreatment;(B) pretreated voice signal is carried out end-point detection by end-point detection program module, judges making an uproar in voice signal
Acoustic frame and speech frame, and the two is separated;(C) noise frame is carried out feature extraction by feature extraction program module, is extracted special
Levy vector;(D) this feature vector is carried out CNN operation by scene Recognition program module, obtains the probability value of each default scene, will be general
The maximum scene of rate value is determined as noise scene;(E) noise reduction process program module selects corresponding gain according to the noise scene
Table G (γk,ξk), and gain table G (γ is chosen by calculating prior weight and posteriori SNRk,ξk) in concrete sound
Enhance parameter, pretreated voice signal is then subjected to speech enhan-cement;(F) tactful coded program module is by enhanced sound
Sound signal is converted into encoded information output.
2. artificial cochlea's sound enhancement method as described in claim 1, it is characterised in that: in step E, gain table G
(γk,ξk) γk, ξkRange take -19dB-20dB, training step is as follows:
(a) local minimum of signals with noise power spectrum is tracked:
(b) voice existing probability: Sr=P (λ, k)/P is calculatedmin(λ, k),
δ (k) is to determine empirical value related with frequency by experiment, and being judged as voice if Sr > δ (k), there are frequency bands, no
It is then grass, that is:
According to the judgment rule of above formula, carry out more new speech existing probability:
P (λ, k)=αpp(λ-1,k)+(1-αp) I (λ, k), wherein αp=0.2;
(c) smoothing constant related with frequency: α is calculateds(λ, k)=αd+(1-αd) p (λ, k), wherein αd=0.85, αs(λ,k)
Value range be αd≤αs(λ,k)≤1;
(d) noise power spectrum: D (λ, k)=α is updateds(λ,k)D(λ-1,k)+(1-αs(λ, k)) Shu Y (λ, k) Shu2, wherein D (λ, k)
It is the estimated value of noise power spectrum, wherein gain, which calculates, uses the least-mean-square error algorithm (LOGSTAS- based on logarithmic spectrum
MMSE), the amplitude spectrum estimator of voice:
Gain function:
Above formula integration type can approximate calculation, i.e.,
3. artificial cochlea's sound enhancement method as claimed in claim 2, it is characterised in that: the calculating formula of the prior weight isThe calculating formula of the posteriori SNR isWherein, λs(k)、λn(k) for voice under k-th frequency band and
The variance of noise, Y are noisy speech.
4. artificial cochlea's sound enhancement method as claimed in claim 3, it is characterised in that: in step, which includes
Framing, adding window, preemphasis and its frequency domain conversion, wherein the adding window uses Hamming window or Hanning window.
5. artificial cochlea's sound enhancement method as claimed in claim 3, it is characterised in that: in stepb, which adopts
With double threshold method, wherein the double threshold method is energy threshold and zero-crossing rate thresholding.
6. artificial cochlea's sound enhancement method as claimed in claim 3, it is characterised in that: in step C, this feature extraction is adopted
With MFCC, FBank or sound spectrograph.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910184264.XA CN109859768A (en) | 2019-03-12 | 2019-03-12 | Artificial cochlea's sound enhancement method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910184264.XA CN109859768A (en) | 2019-03-12 | 2019-03-12 | Artificial cochlea's sound enhancement method |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109859768A true CN109859768A (en) | 2019-06-07 |
Family
ID=66900458
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910184264.XA Withdrawn CN109859768A (en) | 2019-03-12 | 2019-03-12 | Artificial cochlea's sound enhancement method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109859768A (en) |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111144347A (en) * | 2019-12-30 | 2020-05-12 | 腾讯科技(深圳)有限公司 | Data processing method, device, platform and storage medium |
CN111261182A (en) * | 2020-05-07 | 2020-06-09 | 上海力声特医学科技有限公司 | Wind noise suppression method and system suitable for cochlear implant |
CN111341337A (en) * | 2020-05-07 | 2020-06-26 | 上海力声特医学科技有限公司 | Sound noise reduction algorithm and system thereof |
CN112002339A (en) * | 2020-07-22 | 2020-11-27 | 海尔优家智能科技(北京)有限公司 | Voice noise reduction method and device, computer-readable storage medium and electronic device |
Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1861026A (en) * | 2006-05-15 | 2006-11-15 | 清华大学深圳研究生院 | Electronic cochlea language processing method having S parameter control and based-on Chinese language characteristics |
CN101505448A (en) * | 2008-01-21 | 2009-08-12 | 伯纳方股份公司 | A hearing aid adapted to a specific type of voice in an acoustical environment, a method and use |
CN101529929A (en) * | 2006-09-05 | 2009-09-09 | Gn瑞声达A/S | A hearing aid with histogram based sound environment classification |
CN103531204A (en) * | 2013-10-11 | 2014-01-22 | 深港产学研基地 | Voice enhancing method |
CN105390142A (en) * | 2015-12-17 | 2016-03-09 | 广州大学 | Digital hearing aid voice noise elimination method |
US9536540B2 (en) * | 2013-07-19 | 2017-01-03 | Knowles Electronics, Llc | Speech signal separation and synthesis based on auditory scene analysis and speech modeling |
CN109448755A (en) * | 2018-10-30 | 2019-03-08 | 上海力声特医学科技有限公司 | Artificial cochlea's auditory scene recognition methods |
CN109448702A (en) * | 2018-10-30 | 2019-03-08 | 上海力声特医学科技有限公司 | Artificial cochlea's auditory scene recognition methods |
-
2019
- 2019-03-12 CN CN201910184264.XA patent/CN109859768A/en not_active Withdrawn
Patent Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN1861026A (en) * | 2006-05-15 | 2006-11-15 | 清华大学深圳研究生院 | Electronic cochlea language processing method having S parameter control and based-on Chinese language characteristics |
CN101529929A (en) * | 2006-09-05 | 2009-09-09 | Gn瑞声达A/S | A hearing aid with histogram based sound environment classification |
CN101505448A (en) * | 2008-01-21 | 2009-08-12 | 伯纳方股份公司 | A hearing aid adapted to a specific type of voice in an acoustical environment, a method and use |
US9536540B2 (en) * | 2013-07-19 | 2017-01-03 | Knowles Electronics, Llc | Speech signal separation and synthesis based on auditory scene analysis and speech modeling |
CN103531204A (en) * | 2013-10-11 | 2014-01-22 | 深港产学研基地 | Voice enhancing method |
CN105390142A (en) * | 2015-12-17 | 2016-03-09 | 广州大学 | Digital hearing aid voice noise elimination method |
CN109448755A (en) * | 2018-10-30 | 2019-03-08 | 上海力声特医学科技有限公司 | Artificial cochlea's auditory scene recognition methods |
CN109448702A (en) * | 2018-10-30 | 2019-03-08 | 上海力声特医学科技有限公司 | Artificial cochlea's auditory scene recognition methods |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111144347A (en) * | 2019-12-30 | 2020-05-12 | 腾讯科技(深圳)有限公司 | Data processing method, device, platform and storage medium |
CN111261182A (en) * | 2020-05-07 | 2020-06-09 | 上海力声特医学科技有限公司 | Wind noise suppression method and system suitable for cochlear implant |
CN111341337A (en) * | 2020-05-07 | 2020-06-26 | 上海力声特医学科技有限公司 | Sound noise reduction algorithm and system thereof |
CN111341337B (en) * | 2020-05-07 | 2023-03-14 | 上海力声特医学科技有限公司 | Sound noise reduction algorithm and system thereof |
CN112002339A (en) * | 2020-07-22 | 2020-11-27 | 海尔优家智能科技(北京)有限公司 | Voice noise reduction method and device, computer-readable storage medium and electronic device |
CN112002339B (en) * | 2020-07-22 | 2024-01-26 | 海尔优家智能科技(北京)有限公司 | Speech noise reduction method and device, computer-readable storage medium and electronic device |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109859768A (en) | Artificial cochlea's sound enhancement method | |
CN105741849B (en) | The sound enhancement method of phase estimation and human hearing characteristic is merged in digital deaf-aid | |
CN105611477B (en) | The voice enhancement algorithm that depth and range neutral net are combined in digital deaf-aid | |
CN103236260B (en) | Speech recognition system | |
CN107103901B (en) | Artificial cochlea sound scene recognition system and method | |
CN108447495B (en) | Deep learning voice enhancement method based on comprehensive feature set | |
EP3469586B1 (en) | Recursive noise power estimation with noise model adaptation | |
CN109448755A (en) | Artificial cochlea's auditory scene recognition methods | |
CN105261359B (en) | The noise-canceling system and noise-eliminating method of mobile microphone | |
CN108597505B (en) | Voice recognition method and device and terminal equipment | |
CN103778920A (en) | Speech enhancing and frequency response compensation fusion method in digital hearing-aid | |
CN102483926B (en) | System and method for noise reduction in processing speech signals by targeting speech and disregarding noise | |
CN105513605A (en) | Voice enhancement system and method for cellphone microphone | |
CN105448303A (en) | Voice signal processing method and apparatus | |
CN1750124A (en) | Bandwidth extension of band limited audio signals | |
WO2020087716A1 (en) | Auditory scene recognition method for artificial cochlea | |
CN103021405A (en) | Voice signal dynamic feature extraction method based on MUSIC and modulation spectrum filter | |
CN105448302B (en) | A kind of the speech reverberation removing method and system of environment self-adaption | |
CN109151692A (en) | Hearing aid based on deep learning network tests method of completing the square certainly | |
CN112017632A (en) | Automatic conference record generation method | |
CN109671446A (en) | A kind of deep learning sound enhancement method based on absolute hearing threshold | |
US20230226352A1 (en) | Neural Network Audio Scene Classifier for Hearing Implants | |
CN109410919A (en) | A kind of intelligent home control system | |
CN203165457U (en) | Voice acquisition device used for noisy environment | |
Zheng et al. | A noise-robust signal processing strategy for cochlear implants using neural networks |
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 | ||
WW01 | Invention patent application withdrawn after publication |
Application publication date: 20190607 |
|
WW01 | Invention patent application withdrawn after publication |