CN106328160A - Double microphones-based denoising method - Google Patents
Double microphones-based denoising method Download PDFInfo
- Publication number
- CN106328160A CN106328160A CN201510355966.1A CN201510355966A CN106328160A CN 106328160 A CN106328160 A CN 106328160A CN 201510355966 A CN201510355966 A CN 201510355966A CN 106328160 A CN106328160 A CN 106328160A
- Authority
- CN
- China
- Prior art keywords
- signal
- power spectrum
- noise
- frame
- channel
- 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
Landscapes
- Soundproofing, Sound Blocking, And Sound Damping (AREA)
- Circuit For Audible Band Transducer (AREA)
Abstract
A double microphones-based denoising method provided by the present invention comprises a step a of acquiring a first voice signal containing noise and coming from a first microphone via a first channel, acquiring a second voice signal containing noise and coming from a second microphone via a second channel, sampling the first and second voice signals and then carrying out the framing processing to obtain a first channel frame signal and a second channel frame signal; a step b of calculating a correlation coefficient module value of the first and second channel frame signals; a step c of carrying out the frame signal power spectrum calculating separately, and carrying out the noise power spectrum prediction separately; a step d of calculating a prediction signal power spectrum, and calculating the frequency line adjustment gain of the first and second channel frame signals by combining the prediction signal power spectrum and a noise power spectrum, and finally realizing the adaptive filtering on the first and second channel frame signals. According to the technical scheme of the present invention, the prediction accuracy is improved, and a very good background noise suppression effect is realized.
Description
Technical field
The invention belongs to communication field, be particularly based on the diamylose gram noise reduction technology of mobile communication equipment.
Background technology
The appearance of voip technology in recent years so that voice call cost is lower, exchanges more convenient and quicker.The development maked rapid progress along with voice call technology, VoIP terminal is used in our daily life and work exchange frequently.People are more and more higher to the requirement of VoIP voice call quality, and while reducing noise as far as possible, voice signal to reduce the verity of speaker's sound as far as possible.When VoIP terminal is in some important working conferences, user is harsher to the requirement of its noise reduction technology, need to be controlled outside the earshot of human ear by noise while ensureing phonetic hearing impression as far as possible, and in the case of call terminal, still can guarantee that speech quality.
Existing voice de-noising technology mainly has subtractive method of spectrums and automatic adaptation-in.But, both approaches all has some limitations, and subtractive method of spectrums is poor to the tracking effect of noise, and cannot eliminate background noise.And existing automatic adaptation-in uses fixing parameter to be predicted the power spectrum of signal mostly, thus obtain adaptive de-noising wave filter to reach the effect of de-noising.It is short of for the real-time performance of tracking of noise and voice, in the case of noise suppressed is relatively big, and the easy distortion of voice, and when there being preferable voice quality, noise is relatively big, disturbs voice content, causes asaphia clear.Therefore, how to predict power spectrum signal more flexibly, thus compose noise power spectrum and phonetic speech power and make a distinction, utilizing predicts the outcome carries out challenge and the key point that adaptive-filtering is VoIP terminal noise reduction technology to voice signal.
Summary of the invention
Present invention aim at providing a kind of noise-reduction method based on diamylose gram so that the set noise produced due to heat radiation during VoIP terminal operating can not only be eliminated, and background noise can be had good inhibition.
A kind of noise-reduction method based on diamylose gram, mainly comprises the steps that
Step a: first passage collection contains noisy first voice signal from the first mike, second channel gathers from second microphone containing noisy second voice signal, and obtains first passage frame signal and second channel frame signal to carrying out sub-frame processing after the first voice signal and the second speech signal samples respectively；
Step b: calculate first passage frame signal and the coherence factor modulus value of second channel frame signal
；
Step c: carry out the calculating of frame signal power spectrum according to first passage frame signal and second channel frame signal respectively, and carry out noise power spectrum prediction respectively；
Step d: according to prediction signal to noise ratio, coherence factor modulus value, current frame signal power spectrum, former frame noise power spectrum calculate prediction signal power spectrum, and each frequency line combining prediction signal power spectrum, noise power spectrum calculating first passage frame signal and second channel frame signal adjusts gain, finally realize first passage frame signal and the adaptive-filtering of second channel frame signal.
Preferably, in a kind of based on diamylose gram the noise-reduction method that the present invention mentions, above-mentioned coherence factor modulus valueFor:
, wherein,WithBeing respectively the average of the discrete time-domain frame signal of first passage and second channel, computing formula is as follows:
,, in first passage frame signal, the sampled point signal is designated as, in second channel frame signalIndividual sampled point signal is designated as。
Preferably, in this noise-reduction method based on diamylose gram, first passage and second channel frame signal power spectrum are designated as respectively:,, the of first passage frame signalIndividual spectral coefficient is designated as, the of second channel frame signalIndividual spectral coefficient is designated as, the formula of frame signal power spectrum is as follows:
。
Preferably, in this noise-reduction method based on diamylose gram, the present frame noise power spectrum of first passage frame signal and second channel frame signal is designated as respectively,,, the former frame noise power spectrum of first passage frame signal and second channel frame signal is designated as,,, computing formula is as follows:
。
Preferably, in this noise-reduction method based on diamylose gram, it was predicted that the computing formula of power spectrum signal is:,,, wherein,WithThe former frame being respectively first passage and second channel from power spectrum signal, adaptive prediction coefficient is,, it was predicted that signal to noise ratio is:,。
Preferably, in this noise-reduction method based on diamylose gram, the frequency domain in conjunction with described prediction signal power spectrum, noise power spectrum calculating first passage frame signal and second channel frame signal adjusts gain, is respectively as follows:,,, wherein,,,,,For first passage prediction voice signal and second channel prediction voice signal difference signal between the two,WithIt is respectively first passage and the channel parameter of second channel,,It is to improve, according to Wiener filter, the frequency domain adaptive filtering gain obtained.
In order to realize dialogue machine set noise and the suppression of external context noise in VoIP communication process, and voice call is spoken as far as possible close to true people.A kind of based on diamylose gram the noise-reduction method that the present invention provides, has preferable noise reduction in the range of speech range mike 1m, and high to voice restoration degree.
The technical scheme provided due to the present invention is with passage noise when two passage dependencys as basis for estimation to the prediction of power spectrum signal, it can provide preferably filtering at the signal segment that noise contribution is more more accurately, and account for leading signal segment at phonetic element and keep speech intelligibility as far as possible, this makes the voice in communication process maintain higher reduction degree.Additionally, the scheme that the present invention provides uses buffer to store the real time data calculated needed for prediction signal power spectrum and prediction noise power spectrum, algorithm complex is low so that the voice de-noising module used time is shorter, it is ensured that the real-time of voice call.
Accompanying drawing explanation
Fig. 1 be the inventive method from sound collection to noise reduction after output FB(flow block).
Detailed description of the invention
In order to make the purpose of the present invention, technical scheme and advantage clearer, below in conjunction with drawings and Examples, the technical scheme in the embodiment of the present invention is carried out clear, complete description.Should be appreciated that specific embodiment described herein, only in order to explain the present invention, is not intended to limit the present invention.Based on the embodiment in the present invention, the every other embodiment that those skilled in the art is obtained broadly falls into the scope of protection of the invention.
Its general principles is: the form using spectrum subtraction and self-adapted noise elimination to combine reaches the de-noising effect to VoIP terminal.Its signal gathered according to two mikes, obtains predicting noise power spectrum and prediction signal power spectrum, and designs sef-adapting filter according to it, be filtered two sound channels respectively, and regulates parameter and make noise reduction reach optimal.The present invention to the prediction of noise power spectrum in communication process be with call starting section gather signal as foundation.Current signal power spectrum and adjacent former frame power spectrum signal weighted average are obtained by prediction signal power spectrum according to adaptive prediction coefficient.Power spectrum signal predictive coefficient is that the correlation coefficient modulus value adjustment of the signal to noise ratio according to signal and two channels obtains.
Embodiments provide a kind of method that the VoIP communicator of double track collection sound is carried out noise reduction, the principle of certain dependency and independence is there is in the method according to the acoustical signal of two channel acquisition, the foundation adjusted as the predictive coefficient of current signal power spectrum prediction using the signal to noise ratio of signal, thus realize the power spectrum of two paths of signals is realized adaptive prediction, noise segment is tracked prediction simultaneously, improves the anti-acoustic capability of noise filter.
Fig. 1 be the inventive method from sound collection to noise reduction after output FB(flow block).As shown in the drawing, the main contents of the present invention include below scheme step:
Step 1: No. 1 mike collection to VoIP terminal contains noisy voice signal, is designated as, No. 2 microphone apparatus collections, containing noisy voice signal, are designated as。
Step 2: rightWithCarrying out sub-frame processing respectively, each frame frame length is fixed.
Step 3: calculate the coherence factor modulus value of 1,2 passage frame signals。
Step 4: to 1,2 passage frame signals carry out fft(discrete Fourier transform after adding rectangular window) conversion.
Step 5: according to frame signal fft coefficient, calculates signals and associated noises power spectrum.
Step 6: to 1, the noise segment of 2 passage frame signals carries out noise power spectrum prediction.
Step 7: obtain 1 according to predictive coefficient, the prediction power spectrum of 2 passage frame signals.
Wherein, the predictive coefficient of power spectrum obtains according to signal to noise ratio, has the function of self-adaptative adjustment signal estimation.And in order to more be accurately obtained stable noise signal, this programme is use fixing predictive coefficient that noise segment signal is tracked prediction.
Step 8: according to the prediction signal power spectrum obtained, noise power spectrum, each frequency line calculating 1,2 passage frame signals adjusts gain.
Step 9: adjust gain according to the frame signal frequency domain that step 8 obtains, it is achieved the adaptive-filtering to frame signal, filtered signal is carried out fft inverse transformation, obtains the clear voice signal after noise reduction filtering.
Specific embodiments is as described below:
In above-mentioned steps 1, No. 1 mike built-in in VoIP terminal and No. 2 mikes start to gather signal from call successful connection, and the signal of collection is containing noisy voice signal.Institute's Noise is mainly made up of the outside noise two parts in the set noise produced during VoIP terminal operating and call environment.1 passage Noise voice signal of No. 1 mike collection is converted to 1 channel discrete numeral Noise voice signal by analog-to-digital conversion module, is designated as.2 passage signals and associated noises of No. 2 mike collections are converted to 2 channel discrete numeral noisy speech signal through analog-to-digital conversion module, are designated as。
Wherein, No. 1 mike and No. 2 mikes lay respectively at left side and the right side of VoIP terminal inner, in order to collected sound signal from different directions, it is to avoid in communication process, directivity occurs in acoustical signal.No. 1 and No. 2 mikes are placed close together.Owing to position and the direction of two mike collected sound signals are variant, the acoustical signal energy that therefore both gather has larger difference, and owing to two mikes are placed close together, the noise signal that both gather has more relevant portion.
Step 2: to gatherWithDiscrete digital signal, is that a frame carries out framing with 160 discrete voice signaling points, in order to it is carried out real-time adaptive noise reduction process.By in 1 passage frame signalIndividual sampled point signal is designated as, in 2 passage frame signalsIndividual sampled point signal is designated as。
Wherein, the beginning of discrete time-domain frame signal adaptive noise reduction is not formally conversed as mark with VoIP terminal, but be successfully established the mark into starting with call between VoIP terminal.So can increase the accuracy of noise tracking prediction, and avoid the appearance of noise bigger situation when call starts.
Step 3:1, the phonetic element irrelevance in 2 channel signals is relatively big, and noise contribution dependency is relatively big, and this programme designs 1, the coherence factor modulus value of 2 passage frame signalsThe predictive coefficient of Automatic adjusument 1,2 Channel Prediction power spectrum signal is carried out as one of reference factor.This programme design is calculated according to classical two paths of signals Calculation of correlation factor formula:
Wherein,WithBeing respectively the average of 1,2 channel discrete time domain frame signals, computing formula is as follows:
,。
Step 4: to 1, carries out fft conversion after 2 channel discrete time domain frame signal windowed function, obtains the spectral coefficient of two passage frame signals, by the of 1 passage frame signalIndividual spectral coefficient is designated as, the of 2 passage frame signalsIndividual spectral coefficient is designated as:。
Wherein, window function selectivity is more, mainly has rectangular window, Hamming window, Hanning windows etc., according to actual needs and effect, can select different windowed function, in present example, rectangular window has more preferable effect, and therefore the present invention designs selection rectangular window and converts windowed function as fft.
Step 5: 1 passage obtained according to step 4,2 passage frame frequency spectral coefficients, calculate 1 passage, 2 passage frame signal power spectrum, be designated as respectively,, formula is as follows:
,。
Wherein, frame signal power spectrum be frame signal discrete Fourier transform amplitude square.General spectral coefficient is made up of real part and imaginary part two parts, to any spectral coefficientAll it is represented by,For its real part part,For its imaginary part part, then its corresponding spectra calculation formula is represented by:。
The present invention program is to 1, and the signals and associated noises power spectrum of 2 passages and noise power spectrum are predicted respectively, and obtain, according to predicting the outcome, the adaptive frequency domain gain coefficient being filtered current frame signal.For the real-time phonetic element updated in communication process and noise contribution so that predict the outcome more accurate, it was predicted that obtaining according to previous consecutive frame and the power spectrum signal of present frame and noise power spectrum of power spectrum signal and noise power spectrum.The present invention program arranges frame signal power spectrum buffer to store the power spectrum signal of adjacent former frame, arranges noise power spectrum buffer to store the noise power spectrum of adjacent former frame.1,2 channel signal power spectrum of the previous consecutive frame of storage in described power spectrum buffer are designated as respectively,, the adjacent former frame noise power spectrum of storage in noise power spectrum buffer is designated as。
Step 6: the power spectrum signal according to the noise power spectrum in noise power spectrum buffer with by the calculated present frame of step 5 is calculated the noise power spectrum of present frame.The noise power spectrum of 1,2 passage present frames is designated as respectively, computing formula is as follows:
,
。
Wherein, for the first frame signal, deposit in its noise buffer is initialized data, is full 0 data under normal circumstances, and the data in noise buffer used in frame afterwards are then the noise power spectrum of its previous consecutive frame.Having of short duration interval time owing to being typically successfully established between formal call at VoIP terminal call, during this, two channel signals of mike collection are substantially noise signal composition, and this time interval is generally about 2s.This section of noise interval is assumed to be 0.5s by present invention design, owing in case of the present invention, the sample rate of voice signal is 16khz, therefore the interval of 0.5 is the discrete digital signal of front 50 frames, and it is essentially noise signal, therefore is tracked predicting to noise signal at front 50 frames.In order to avoid voice signal being mistaken for noise contribution, noise prediction result is produced mistake interference, signal after 50 frames is no longer carried out noise tracking prediction, but the steady state noise that obtains of former 50 frames predictions as frame noise signal to calculate prediction signal power spectrum.
Step 7: according to prediction signal to noise ratio, the correlation coefficient modulus value calculated in step 3, in step 5, the current frame signal power spectrum of calculating, the data in power spectrum signal buffer, calculate prediction signal power spectrum.Prediction signal to noise ratio is the prediction signal power spectrum according to the wherein adjacent former frame signal of 1 passage and noise power spectrum calculating prediction signal to noise ratio, is designated as, selection 1 passage not affected in the present invention program calculates.Prediction signal power spectrum is based on adaptive prediction coefficient and computes weighted the data in power spectrum signal and power spectrum signal buffer and obtain.Adaptive prediction coefficient is designated as, its be based on andCarry out what Automatic adjusument obtained.The prediction signal power spectrum of 1,2 passages is designated as.Owing to the calculating of prediction signal to noise ratio needs the prediction signal power spectrum according to adjacent former frame and noise power spectrum, therefore the present invention arranges prediction signal power spectrum buffer, 1 will wherein stored, and the prediction signal power spectrum of the 2 adjacent former frame of passage is designated as。
7-a. calculates according to the data in prediction signal power spectrum caching and the data in noise power spectrum buffer and predicts signal to noise ratio:
,。
The value of adaptive prediction coefficient is calculated by 7-b. according to the size of prediction signal to noise ratio:
,。
7-c.
In order to avoid adaptive prediction coefficient is excessive or too small impacts prediction signal power spectrum, its span is adjusted so that it is span controls at 0-0.95:
。
7-d.
Prediction signal power spectrum is obtained according to adaptive prediction coefficient calculations:
,
,。
Wherein, for the first frame signal, power spectrum signal buffer, it was predicted that the data in power spectrum signal buffer are all initialized data, being full 0 data under normal circumstances, for other frame signals, wherein deposit is then the data of corresponding adjacent former frame.To three i.e. noise power spectrum buffers of buffer, power spectrum signal buffer, it was predicted that the data in power spectrum signal buffer carry out real-time update described in case of the present invention after being calculated the prediction signal power spectrum of present frame.
Step 8: the prediction signal power spectrum that the noise power spectrum obtained according to above-mentioned steps 6 and step 7 obtain, calculates the frequency coefficient adaptive gain regulating of current frame signal.By the of 1 passage frame signalThe gain that root frequency line is corresponding is designated as, the of 2 passage frame signalsThe gain that root frequency line is corresponding is designated as, its computing formula is as follows:
,。
Wherein,
,,。
Wherein,It is the difference signal of two Channel Prediction voice signals, andWithIt is respectively the channel parameter of 1,2 passages,,It is to improve, according to Wiener filter, the frequency domain adaptive filtering gain obtained.
Step 9: according to the signal frequency domain regulation gain obtained in step 8 to 1, the frame signal of 2 passages is filtered processing, change filtered frequency coefficient contravariant into time-domain signal again, i.e. obtain 1 after noise-reduction method noise reduction process based on double track power spectrum adaptive prediction, 2 channel speech signals, are designated as the frame signal after 1 passage noise reduction, the frame signal after 2 passage noise reductions is designated as, wherein,For inversefouriertransform:
。
The foregoing is only presently preferred embodiments of the present invention, not in order to limit the present invention, all any amendment, equivalent and improvement etc. made within the spirit and principles in the present invention, should be included within the scope of the present invention.
Claims (10)
1. a noise-reduction method based on diamylose gram, it is characterised in that mainly comprise the steps that
Step a: first passage collection contains noisy first voice signal from the first mike, second channel gathers from second microphone containing noisy second voice signal, and obtains first passage frame signal and second channel frame signal to carrying out sub-frame processing after described first voice signal and the second speech signal samples respectively；
Step b: calculate the coherence factor modulus value of described first passage frame signal and second channel frame signal
；
Step c: carry out the calculating of frame signal power spectrum according to described first passage frame signal and described second channel frame signal respectively, and carry out noise power spectrum prediction respectively；
Step d: according to prediction signal to noise ratio, described coherence factor modulus value, current frame signal power spectrum, former frame noise power spectrum calculate prediction signal power spectrum, and each frequency line combining described prediction signal power spectrum, the described noise power spectrum described first passage frame signal of calculating and described second channel frame signal adjusts gain, finally realize described first passage frame signal and the adaptive-filtering of described second channel frame signal.
Noise-reduction method the most according to claim 1, it is characterised in that described coherence factor modulus valueFor:
, wherein, describedWith describedBeing respectively the average of the discrete time-domain frame signal of described first passage and described second channel, computing formula is as follows:
,, in described first passage frame signalIndividual sampled point signal is designated as, in described second channel frame signalIndividual sampled point signal is designated as。
Noise-reduction method the most according to claim 2, it is characterised in that described first passage and described second channel frame signal power spectrum are designated as respectively:,, the of described first passage frame signalIndividual spectral coefficient is designated as, the of described second channel frame signalIndividual spectral coefficient is designated as, the formula of frame signal power spectrum is as follows:
,。
Noise-reduction method the most according to claim 3, it is characterised in that the present frame noise power spectrum of described first passage frame signal and described second channel frame signal is designated as respectively,,, the former frame noise power spectrum of described first passage frame signal and described second channel frame signal is designated as,,, computing formula is as follows:
。
Noise-reduction method the most according to claim 4, it is characterised in that the computing formula of described prediction signal power spectrum is:,,, wherein, describedWith describedRespectively the former frame of described first passage and described second channel is from power spectrum signal, describedWith describedBeing respectively described first passage and the former frame prediction signal power spectrum of described second channel, adaptive prediction coefficient is,, described prediction signal to noise ratio is:,。
Noise-reduction method the most according to claim 5, it is characterised in that the frequency domain combining described prediction signal power spectrum, the described noise power spectrum described first passage frame signal of calculating and described second channel frame signal adjusts gain, is respectively as follows:,,, wherein,,,,, describedFor described first passage prediction voice signal and described second channel prediction voice signal difference signal between the two, describedWith describedIt is respectively described first passage and the channel parameter of described second channel,,It is to improve, according to Wiener filter, the frequency domain adaptive filtering gain obtained.
Noise-reduction method the most according to claim 6, it is characterised in that during to described first voice signal and the second speech signal samples with sample rate 16khz, can be tracked prediction at front 50 frames to noise signal.
Noise-reduction method the most according to claim 7, it is characterised in that described former frame noise power spectrum is stored in noise power spectrum buffer；Described former frame power spectrum signal is stored in power spectrum signal buffer；The prediction signal power spectrum of described former frame is stored in prediction signal power spectrum buffer.
Noise-reduction method the most according to claim 8, it is characterized in that, adjust gain according to described frequency domain to be filtered described first passage frame signal and described second channel frame signal respectively processing, change filtered frequency coefficient contravariant into time-domain signal the most again, i.e. obtain the first passage voice signal after noise reduction process and second channel voice signal.
Noise-reduction method the most according to claim 9, it is characterised in that the data in described noise power spectrum buffer, described power spectrum signal buffer and described prediction signal power spectrum buffer carry out real-time update.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510355966.1A CN106328160B (en) | 2015-06-25 | 2015-06-25 | Noise reduction method based on double microphones |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510355966.1A CN106328160B (en) | 2015-06-25 | 2015-06-25 | Noise reduction method based on double microphones |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106328160A true CN106328160A (en) | 2017-01-11 |
| CN106328160B CN106328160B (en) | 2021-03-02 |
Family
ID=57729183
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510355966.1A Active CN106328160B (en) | 2015-06-25 | 2015-06-25 | Noise reduction method based on double microphones |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106328160B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107371079A (en) * | 2017-04-17 | 2017-11-21 | 恒玄科技（上海）有限公司 | The diamylose gram noise reduction system and noise-reduction method of a kind of earphone |
| CN108109631A (en) * | 2017-02-10 | 2018-06-01 | 深圳市启元数码科技有限公司 | A kind of small size dual microphone voice collecting noise reduction module and its noise-reduction method |
| CN109327755A (en) * | 2018-08-20 | 2019-02-12 | 深圳信息职业技术学院 | A kind of cochlear implant and noise remove method |
| CN109616135A (en) * | 2018-11-14 | 2019-04-12 | 腾讯音乐娱乐科技（深圳）有限公司 | Audio-frequency processing method, device and storage medium |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPH0690493A (en) * | 1992-09-09 | 1994-03-29 | Sony Corp | Voice input device and image pickup device using this |
| WO2007059255A1 (en) * | 2005-11-17 | 2007-05-24 | Mh Acoustics, Llc | Dual-microphone spatial noise suppression |
| EP1931169A1 (en) * | 2005-09-02 | 2008-06-11 | Japan Advanced Institute of Science and Technology | Post filter for microphone array |
| US20100046770A1 (en) * | 2008-08-22 | 2010-02-25 | Qualcomm Incorporated | Systems, methods, and apparatus for detection of uncorrelated component |
| EP2770750A1 (en) * | 2013-02-25 | 2014-08-27 | Spreadtrum Communications (Shanghai) Co., Ltd. | Detecting and switching between noise reduction modes in multi-microphone mobile devices |
-
2015
- 2015-06-25 CN CN201510355966.1A patent/CN106328160B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JPH0690493A (en) * | 1992-09-09 | 1994-03-29 | Sony Corp | Voice input device and image pickup device using this |
| EP1931169A1 (en) * | 2005-09-02 | 2008-06-11 | Japan Advanced Institute of Science and Technology | Post filter for microphone array |
| WO2007059255A1 (en) * | 2005-11-17 | 2007-05-24 | Mh Acoustics, Llc | Dual-microphone spatial noise suppression |
| US20100046770A1 (en) * | 2008-08-22 | 2010-02-25 | Qualcomm Incorporated | Systems, methods, and apparatus for detection of uncorrelated component |
| EP2770750A1 (en) * | 2013-02-25 | 2014-08-27 | Spreadtrum Communications (Shanghai) Co., Ltd. | Detecting and switching between noise reduction modes in multi-microphone mobile devices |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108109631A (en) * | 2017-02-10 | 2018-06-01 | 深圳市启元数码科技有限公司 | A kind of small size dual microphone voice collecting noise reduction module and its noise-reduction method |
| CN107371079A (en) * | 2017-04-17 | 2017-11-21 | 恒玄科技（上海）有限公司 | The diamylose gram noise reduction system and noise-reduction method of a kind of earphone |
| CN109327755A (en) * | 2018-08-20 | 2019-02-12 | 深圳信息职业技术学院 | A kind of cochlear implant and noise remove method |
| CN109616135A (en) * | 2018-11-14 | 2019-04-12 | 腾讯音乐娱乐科技（深圳）有限公司 | Audio-frequency processing method, device and storage medium |
| CN109616135B (en) * | 2018-11-14 | 2021-08-03 | 腾讯音乐娱乐科技（深圳）有限公司 | Audio processing method, device and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106328160B (en) | 2021-03-02 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108831499B (en) | Speech enhancement method using speech existence probability | |
| JP5542952B2 (en) | Microphone array noise reduction control method and apparatus | |
| US8521530B1 (en) | System and method for enhancing a monaural audio signal | |
| CN111418010A (en) | Multi-microphone noise reduction method and device and terminal equipment | |
| US8831936B2 (en) | Systems, methods, apparatus, and computer program products for speech signal processing using spectral contrast enhancement | |
| JP5300861B2 (en) | Noise suppressor | |
| US9113241B2 (en) | Noise removing apparatus and noise removing method | |
| JP5153886B2 (en) | Noise suppression device and speech decoding device | |
| US9854368B2 (en) | Method of operating a hearing aid system and a hearing aid system | |
| US20140025374A1 (en) | Speech enhancement to improve speech intelligibility and automatic speech recognition | |
| WO2012158156A1 (en) | Noise supression method and apparatus using multiple feature modeling for speech/noise likelihood | |
| KR20130108063A (en) | Multi-microphone robust noise suppression | |
| JP2014232331A (en) | System and method for adaptive intelligent noise suppression | |
| CN108899052B (en) | Parkinson speech enhancement method based on multi-band spectral subtraction | |
| CN106328160A (en) | Double microphones-based denoising method | |
| AU2009203194A1 (en) | Noise spectrum tracking in noisy acoustical signals | |
| CN102074245A (en) | Dual-microphone-based speech enhancement device and speech enhancement method | |
| US11128954B2 (en) | Method and electronic device for managing loudness of audio signal | |
| JP2008216720A (en) | Signal processing method, device, and program | |
| CN106340303A (en) | Speech denoising method based on time frequency domain | |
| Jeub et al. | Robust dual-channel noise power spectral density estimation | |
| CN105702262A (en) | Headset double-microphone voice enhancement method | |
| CN103813251A (en) | Hearing-aid denoising device and method allowable for adjusting denoising degree | |
| Jin et al. | Multi-channel noise reduction for hands-free voice communication on mobile phones | |
| JP2013168856A (en) | Noise reduction device, audio input device, radio communication device, noise reduction method and noise reduction program |
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 |