CN107785028B - Voice noise reduction method and device based on signal autocorrelation - Google Patents
Voice noise reduction method and device based on signal autocorrelation Download PDFInfo
- Publication number
- CN107785028B CN107785028B CN201610725981.5A CN201610725981A CN107785028B CN 107785028 B CN107785028 B CN 107785028B CN 201610725981 A CN201610725981 A CN 201610725981A CN 107785028 B CN107785028 B CN 107785028B
- Authority
- CN
- China
- Prior art keywords
- signal
- module
- voice
- autocorrelation
- noise reduction
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 20
- 238000009432 framing Methods 0.000 claims abstract description 21
- 238000006243 chemical reaction Methods 0.000 claims abstract description 19
- 238000005311 autocorrelation function Methods 0.000 claims abstract description 17
- 238000004458 analytical method Methods 0.000 claims abstract description 16
- 238000012952 Resampling Methods 0.000 claims abstract description 15
- 238000005215 recombination Methods 0.000 claims abstract description 15
- 230000006798 recombination Effects 0.000 claims abstract description 15
- 238000005070 sampling Methods 0.000 claims description 8
- 230000009466 transformation Effects 0.000 claims 1
- 238000004891 communication Methods 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 3
- 238000001914 filtration Methods 0.000 description 2
- 230000006870 function Effects 0.000 description 2
- 230000003595 spectral effect Effects 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 210000005069 ears Anatomy 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
- 238000001228 spectrum Methods 0.000 description 1
- 230000001629 suppression Effects 0.000 description 1
- 230000001052 transient effect Effects 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS TECHNIQUES OR SPEECH SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING TECHNIQUES; SPEECH OR AUDIO CODING OR DECODING
- G10L21/00—Speech or voice signal processing techniques to produce another audible or non-audible signal, e.g. visual or tactile, in order to modify its quality or its intelligibility
- G10L21/02—Speech enhancement, e.g. noise reduction or echo cancellation
- G10L21/0208—Noise filtering
- G10L21/0216—Noise filtering characterised by the method used for estimating noise
- G10L21/0232—Processing in the frequency domain
Landscapes
- Engineering & Computer Science (AREA)
- Computational Linguistics (AREA)
- Quality & Reliability (AREA)
- Signal Processing (AREA)
- Health & Medical Sciences (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Human Computer Interaction (AREA)
- Physics & Mathematics (AREA)
- Acoustics & Sound (AREA)
- Multimedia (AREA)
- Noise Elimination (AREA)
Abstract
The invention relates to a voice noise reduction method and a voice noise reduction device based on signal autocorrelation, which comprise the following steps: 1) the voice receiving module collects a noise-containing analog voice signal; 2) the A/D conversion module converts the analog signal into a digital signal; 3) the voice framing module frames the digital signal; 4) the autocorrelation denoising module carries out autocorrelation operation on the framed signal; 5) the FFT analysis module performs FFT analysis on the autocorrelation function of each frame of noisy signals in a time delay domain; 6) the threshold denoising module performs threshold denoising processing on the FFT; 7) the resampling and phase pairing module resamples the frequency domain autocorrelation signal and pairs the frequency domain autocorrelation signal with the phase of the noisy signal; 8) the inverse FFT module carries out inverse FFT by using the processed frequency domain amplitude and phase and returns the signal to the time domain; 9) and the frame recombination module recombines the denoised frame signals to obtain the denoised voice signals. The device can obviously improve the speech definition in a strong noise environment with a signal-to-noise ratio of-20 dB.
Description
Technical Field
The present invention relates to noise reduction technology in the field of speech recognition, and in particular, to a speech noise reduction method and apparatus based on signal autocorrelation.
Background
Voice noise reduction is mainly applied to voice recognition in man-machine conversation and the field of voice communication (including traditional wired communication, wireless communication and network communication). The voice noise greatly reduces the capability of intelligent equipment and human ears to correctly recognize the voice, and the voice noise reduction enhances the recognition rate of the voice in a noise environment. It is therefore also desirable to use speech noise reduction techniques in man-machine dialogues and communication devices that enhance the accuracy with which intelligent devices recognize and execute speech commands, reduce human fatigue from recognizing noisy speech, and improve speech intelligibility.
The conventional speech noise reduction methods are mainly as follows: noise cancellation, harmonic enhancement, speech model enhancement, short-term spectrum enhancement, and filtering. However, due to the complexity and non-stationarity of the speech signal, these conventional noise reduction methods are difficult to reduce noise of a signal with a signal-to-noise ratio lower than 10dB, and perform effective recognition, especially, the suppression effect on broadband noise, transient noise, and strong noise is not satisfactory.
The invention develops a voice noise reduction technology under a strong noise environment based on the pronunciation characteristics of Chinese. The Chinese language is distinguished from the phonation of the alphabetic writing and has the characteristics that: the unvoiced sound must be aided by voiced sound, so the frequency characteristics of the unvoiced sound are obvious, unlike the continuous unvoiced sound of the alphabetic writing, the unvoiced sound has no obvious time domain and frequency domain characteristics during the sound production, and is very similar to white noise. By utilizing the characteristic, after the autocorrelation operation is carried out on the Chinese speech signal, the autocorrelation value of the speech signal with the frequency characteristic is larger, so that the autocorrelation value is enhanced and retained, and the autocorrelation value of the noise signal with the weaker frequency characteristic is smaller and is restrained. In the FFT-transformed frequency domain of the autocorrelation function, it appears that the frequency amplitude of the speech signal is large, and the frequency amplitude of the noise is small. After threshold processing is carried out on the frequency signal, denoising can be further carried out, and denoising of the voice signal is achieved.
Disclosure of Invention
The invention aims to overcome the defects of the existing voice noise reduction technology, and provides a voice noise reduction method and a voice noise reduction device which are accurate, rapid and high in automation degree.
The purpose of the invention can be realized by the following technical scheme:
a voice noise reduction method based on signal autocorrelation comprises the following steps:
1) the voice receiving module receives and collects a voice signal containing noise through a microphone and transmits the voice signal to the A/D conversion module;
2) the A/D conversion module converts the collected analog voice signals into digital signals and transmits the digital signals to the voice framing module;
3) the voice framing module frames the voice digital signal, the time length is 20-30 ms, and then the framed signal is transmitted to the self-correlation denoising module;
4) the autocorrelation denoising module performs autocorrelation operation on the framed signal to realize preliminary denoising, and then transmits the signal to the FFT analysis module;
5) the FFT analysis module performs FFT analysis on the autocorrelation function of each frame of noisy signals in a time delay domain, expands the autocorrelation function into a function in a frequency domain, and then transmits a series of amplitudes in the frequency domain to the threshold noise reduction module;
6) the threshold noise reduction module sets the frequency component amplitude value lower than the threshold value to zero by setting the threshold value to realize secondary noise reduction, and transmits the processed amplitude value to the resampling and phase matching module;
7) the resampling and phase matching module resamples the processed amplitude, takes a half of the number of points, matches the resampled signal amplitude with the phase of the noise-containing signal, and then transmits the combination of the matched amplitude and phase to the inverse FFT conversion module;
8) the inverse FFT conversion module performs inverse FFT conversion on the resampled amplitude and the matched noise-containing signal phase, returns the signal to a time domain, and then transmits the signal in the time domain to the frame recombination module;
9) and the frame recombination module performs frame recombination on the signals to form a complete denoised voice signal so as to complete the denoising process.
The working parameters of the voice receiving module comprise a signal acquisition channel, sampling frequency and acquisition signal duration;
the working parameters of the voice framing module comprise signal framing duration;
the working parameters of the threshold noise reduction module comprise a noise reduction threshold.
A device of a voice noise reduction method based on signal autocorrelation is characterized by comprising a voice receiving module, an A/D conversion module, a voice framing module, an autocorrelation denoising module, an FFT analysis module, a threshold denoising module, a resampling and phase pairing module, an inverse FFT conversion module and a frame recombination module.
The voice receiving module comprises a signal acquisition channel, a sampling frequency and a signal acquisition duration, and sends acquired voice signals to the voice framing module.
The voice framing module needs to set the signal framing duration;
the autocorrelation denoising module performs autocorrelation operation on the signal, and the short-time autocorrelation function of the speech signal s (n) with limited energy is defined as:
m is the number of time delay points, and n is the number of signal duration points.
The FFT analysis module is used for carrying out fast Fourier transform on the autocorrelation function to obtain the frequency domain representation of the autocorrelation function.
The threshold noise reduction module resets the frequency amplitude lower than the set threshold to zero, and keeps the frequency amplitude higher than the threshold unchanged to obtain a group of new Fourier transform coefficients, and sends the new Fourier transform coefficients to the resampling and phase matching module.
The resampling and phase matching module performs point-isolated sampling on the Fourier transform coefficient, then matches the Fourier transform coefficient with the phase of the acquired signal, and combines and sends the Fourier transform coefficient and the phase of the acquired signal to the inverse FFT module;
the inverse FFT module carries out inverse FFT by using the resampled amplitude and the matched phase to obtain a denoised time domain signal and then sends the denoised time domain signal to the frame recombination module;
and the frame recombination module is used for recombining each frame signal after the denoising processing to obtain a complete speech signal after the denoising processing.
The voice signal s (t) collected by the voice receiving module, besides the useful voice signal v (t), also has the noise n (t) in the environment, and makes autocorrelation operation on the signal, i.e. making said signal undergo the process of filtering treatment
s(t)*s(t)=[v(t)+n(t)]*[v(t)+n(t)]
Assuming that the Fourier transforms (FFT) of s (t), v (t), n (t) are S (f), V (f), N (f), respectively, the Fourier transform of the autocorrelation operation of s (t) can be represented as:
F[s(t)*s(t)]=S(f)·S*(f)=[V(f)+N(f)]·[V*(f)+N*(f)]
=V(f)·V*(f)+V(f)·N*(f)+N(f)·V*(f)+N(f)·N*(f)
≈V(f)*V*(f)+Z(f)
i.e. V (t) the amplitude of the autocorrelation in the frequency domain | V (f). V*(f) And | is the square of the amplitude of the original signal in the frequency domain. The frequency domain of the autocorrelation function also contains the frequency component z (f) of the noise signal.
The speech signal can be approximately regarded as a quasi-steady-state signal with remarkable frequency characteristics in a short time, and the noise is a random signal with unstable amplitude, frequency and phase. Through the autocorrelation processing, the voice signal is preserved because the signals at different times have correlation. While noise is suppressed because of the small signal correlation at different times. The correlation operation process can play a role of preliminary noise reduction.
Because the frequency of the voice is concentrated in a limited narrow band, the amplitude of the voice in the frequency domain is large, the frequency band of the noise is wide, the voice is distributed on the whole frequency domain, and the amplitude of the voice is low, so that the frequency component lower than a certain threshold value is used as the noise to be processed, the amplitude of the voice is zero, most of Z (f) is eliminated, and the denoising effect can be realized again.
The frequency domain autocorrelation function after two times of denoising is used for squaring the frequency domain amplitude, so the spectral characteristic of the frequency domain autocorrelation function is very close to the spectral characteristic of a clean voice signal, the point number of the autocorrelation calculation V (f) and V (f) in the frequency domain is two times 2N of the point number V (f), half of the point number N is taken for resampling, and then the resampled signal V' (f) is matched with the phase of a noise-containing signal.
Drawings
FIG. 1 is a schematic structural view of the present invention;
fig. 2 is a time domain waveform diagram of each signal.
Detailed Description
The invention is described in detail below with reference to the figures and specific embodiments.
Examples
As shown in fig. 1, the present invention firstly receives and collects a speech signal polluted by noise, and then performs denoising processing on the speech signal based on a signal autocorrelation technique to obtain a speech signal with high signal-to-noise ratio and greatly improved intelligibility. The denoising device comprises a voice receiving module 1, an A/D conversion module 2, a voice framing module 3, an autocorrelation denoising module 4, an FFT analysis module 5, a threshold denoising module 6, a resampling and phase pairing module 7, an inverse FFT conversion module 8 and a frame recombination module 9.
The voice receiving module receives and collects a voice signal s (t) (v) (t) + n (t) containing noise through a microphone and transmits the voice signal s (t) (v) (t) + n (t) containing noise to the A/D conversion module; the A/D conversion module converts the collected analog voice signals into digital signals and transmits the digital signals to the voice framing module; the voice framing module frames the voice digital signal, the time length is 20-30 ms, and then the framed signal is transmitted to the self-correlation denoising module; the autocorrelation denoising module performs autocorrelation operation on the framed signal to realize preliminary denoising, and then transmits the signal to the FFT analysis module; the FFT analysis module performs FFT analysis on an autocorrelation function s (t) of each frame of noisy signals s (t) in a delay domain, expands the autocorrelation function s (f) of each frame of noisy signals into a function S (f) in a frequency domain, and then transmits a series of amplitudes in the frequency domain to the threshold noise reduction module; the threshold processing is realized by setting a threshold, setting the frequency component lower than the threshold to zero, realizing secondary denoising, and transmitting the processed amplitude combination to a resampling and phase pairing module; the resampling and phase matching module resamples the processed amplitude, takes half of the point number N, matches the resampled signal V' (f) with the phase of the noise-containing signal, and then transmits the combination of the matched amplitude and phase to the inverse FFT conversion module. And the inverse FFT module performs inverse FFT on the V' (f) and the phase of the matched noise-containing signal, returns the signal to a time domain and then transmits the signal in the time domain to the frame recombination module. And the frame recombination module performs frame recombination on the signals to form a complete denoised voice signal so as to complete the denoising process.
The denoising process of the present invention is illustrated below:
a group of pure voice signals are respectively collected under the conditions of no noise interference, being submerged by white noise and signal-to-noise ratio of-20 dB, the sampling rate is 44100, the time length of the collected signals is 0.55s, the voice framing module is 2 (the signals are shorter), and the threshold value noise reduction module sets the threshold value to be 1500 (frequency components smaller than 1500 are all set to zero). FIG. 2 shows a comparison time domain waveform of a clean speech signal, a noisy signal and a denoised signal.
Comparing each oscillogram in fig. 2, it can be seen that after the denoising processing, the signal-to-noise ratio of the denoised signal is greatly improved compared with the noise-added signal, and the denoised signal has better consistency with the original voice signal. Compared with an audio file, the signal speech after being interfered by the noise cannot be identified, but the voice signal after being subjected to denoising processing can be clearly identified.
Claims (10)
1. A speech noise reduction method based on signal autocorrelation is characterized by comprising the following steps:
1) the voice receiving module receives and collects a voice signal containing noise through a microphone and transmits the voice signal to the A/D conversion module;
2) the A/D conversion module converts the collected analog voice signals into digital signals and transmits the digital signals to the voice framing module;
3) the voice framing module frames the voice digital signal, the time length is 20-30 ms, and then the framed signal is transmitted to the self-correlation denoising module;
4) the autocorrelation denoising module performs autocorrelation operation on the framed signal to realize preliminary denoising, and then transmits the signal to the FFT analysis module; the autocorrelation denoising module performs autocorrelation operation on the signal, and the short-time autocorrelation function of the speech signal s (n) with limited energy is defined as follows:
m is the number of time delay points, and n is the number of signal duration points;
5) the FFT analysis module performs FFT analysis on the autocorrelation function of each frame of noisy signals in a time delay domain, expands the autocorrelation function into a function in a frequency domain, and then transmits a series of amplitudes in the frequency domain to the threshold noise reduction module;
6) the threshold noise reduction module sets the frequency component amplitude value lower than the threshold value to zero by setting the threshold value to realize secondary noise reduction, and transmits the processed amplitude value to the resampling and phase matching module;
7) the resampling and phase matching module resamples the processed amplitude, takes a half of the number of points, matches the resampled signal amplitude with the phase of the noise-containing signal, and then transmits the combination of the matched amplitude and phase to the inverse FFT conversion module;
8) the inverse FFT conversion module performs inverse FFT conversion on the resampled amplitude and the matched noise-containing signal phase, returns the signal to a time domain, and then transmits the signal in the time domain to the frame recombination module;
9) and the frame recombination module performs frame recombination on the signals to form a complete denoised voice signal so as to complete the denoising process.
2. The method according to claim 1, wherein the operating parameters of the speech receiving module include a signal acquisition channel, a sampling frequency and a signal acquisition duration;
the working parameters of the voice framing module comprise signal framing duration;
the working parameters of the threshold noise reduction module comprise a noise reduction threshold.
3. The apparatus of the speech noise reduction method based on signal autocorrelation as claimed in claim 1, comprising a speech receiving module, an a/D conversion module, a speech framing module, an autocorrelation denoising module, an FFT analysis module, a threshold noise reduction module, a resampling and phase pairing module, an inverse FFT transformation module and a frame recombination module.
4. The device of claim 3, wherein the voice receiving module comprises a signal acquisition channel, a sampling frequency and a signal acquisition duration, and sends the acquired voice signal to the voice framing module.
5. The apparatus of claim 3, wherein the speech framing module is configured to set a signal framing duration.
6. The apparatus of claim 3, wherein the FFT analysis module performs a fast Fourier transform on the autocorrelation function to obtain a frequency domain representation of the autocorrelation function.
7. The apparatus of claim 3, wherein the threshold noise reduction module resets the frequency amplitudes below the set threshold to zero, and keeps the frequency amplitudes above the threshold unchanged to obtain a new set of Fourier transform coefficients, and sends the new set of Fourier transform coefficients to the resampling and phase matching module.
8. The apparatus of claim 3, wherein the resampling and phase matching module performs alternate sampling on the Fourier transform coefficients, matches the alternate sampling with the phase of the collected signal, and sends the combined signal to the inverse FFT module.
9. The apparatus of claim 3, wherein the inverse FFT module performs inverse FFT with the resampled amplitude and the matched phase to obtain a denoised time domain signal, and then sends the denoised time domain signal to the frame reassembly module.
10. The apparatus of claim 3, wherein the frame reconstructing module reconstructs the denoised frame signals to obtain a complete denoised speech signal.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610725981.5A CN107785028B (en) | 2016-08-25 | 2016-08-25 | Voice noise reduction method and device based on signal autocorrelation |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610725981.5A CN107785028B (en) | 2016-08-25 | 2016-08-25 | Voice noise reduction method and device based on signal autocorrelation |
Publications (2)
Publication Number | Publication Date |
---|---|
CN107785028A CN107785028A (en) | 2018-03-09 |
CN107785028B true CN107785028B (en) | 2021-06-18 |
Family
ID=61438771
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610725981.5A Active CN107785028B (en) | 2016-08-25 | 2016-08-25 | Voice noise reduction method and device based on signal autocorrelation |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107785028B (en) |
Families Citing this family (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108986840A (en) * | 2018-04-03 | 2018-12-11 | 五邑大学 | A kind of recognition methods during detecting electroscope to buzzer audio |
CN109285556B (en) * | 2018-09-29 | 2022-05-20 | 阿波罗智联(北京)科技有限公司 | Audio processing method, device, equipment and storage medium |
CN109587089B (en) * | 2018-12-12 | 2021-04-09 | 北航(四川)西部国际创新港科技有限公司 | Method for improving signal identification accuracy of unmanned aerial vehicle |
CN110430316B (en) * | 2019-06-25 | 2021-05-14 | 努比亚技术有限公司 | Voice noise reduction method, mobile terminal and computer readable storage medium |
CN110971763B (en) * | 2019-12-10 | 2021-01-26 | Oppo广东移动通信有限公司 | Arrival reminding method and device, storage medium and electronic equipment |
CN112190280B (en) * | 2020-10-13 | 2022-10-14 | 苏州美糯爱医疗科技有限公司 | Real-time automatic background sound interference cancellation method for electronic stethoscope |
CN112331225B (en) * | 2020-10-26 | 2023-09-26 | 东南大学 | Method and device for assisting hearing in high-noise environment |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO1999030415A2 (en) * | 1997-12-05 | 1999-06-17 | Telefonaktiebolaget Lm Ericsson (Publ) | Noise reduction method and apparatus |
CN2833787Y (en) * | 2005-11-07 | 2006-11-01 | 郑硕钧 | Noise suppression and anti-interference device for voice signals |
CN102117621A (en) * | 2010-01-05 | 2011-07-06 | 吴伟 | Signal denoising method with self correlation coefficient as the criterion |
Family Cites Families (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US5148488A (en) * | 1989-11-17 | 1992-09-15 | Nynex Corporation | Method and filter for enhancing a noisy speech signal |
DE102004005998B3 (en) * | 2004-02-06 | 2005-05-25 | Ruwisch, Dietmar, Dr. | Separating sound signals involves Fourier transformation, inverse transformation using filter function dependent on angle of incidence with maximum at preferred angle and combined with frequency spectrum by multiplication |
-
2016
- 2016-08-25 CN CN201610725981.5A patent/CN107785028B/en active Active
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO1999030415A2 (en) * | 1997-12-05 | 1999-06-17 | Telefonaktiebolaget Lm Ericsson (Publ) | Noise reduction method and apparatus |
CN2833787Y (en) * | 2005-11-07 | 2006-11-01 | 郑硕钧 | Noise suppression and anti-interference device for voice signals |
CN102117621A (en) * | 2010-01-05 | 2011-07-06 | 吴伟 | Signal denoising method with self correlation coefficient as the criterion |
Non-Patent Citations (2)
Title |
---|
Speech Enhancement with Zero Replacement Signal;Iwai Yu等;《IEEE》;20151231;第143-147页 * |
基于车载数字对讲机的语音降噪技术研究;黄春晖;《中国优秀硕士学位论文数据库 信息科技辑》;20081015(第10期);第I136-63页 * |
Also Published As
Publication number | Publication date |
---|---|
CN107785028A (en) | 2018-03-09 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107785028B (en) | Voice noise reduction method and device based on signal autocorrelation | |
CN110600050B (en) | Microphone array voice enhancement method and system based on deep neural network | |
CN109584903A (en) | A kind of multi-person speech separation method based on deep learning | |
Xiao et al. | Normalization of the speech modulation spectra for robust speech recognition | |
CN105390142B (en) | A kind of digital deaf-aid voice noise removing method | |
CN106340303B (en) | A kind of voice de-noising method based on temporal frequency domain | |
CN108198545B (en) | Speech recognition method based on wavelet transformation | |
WO2001033550A1 (en) | Speech parameter compression | |
CN109523999A (en) | A kind of front end processing method and system promoting far field speech recognition | |
CN105719657A (en) | Human voice extracting method and device based on microphone | |
CN110808057A (en) | Voice enhancement method for generating confrontation network based on constraint naive | |
CN104240717B (en) | Voice enhancement method based on combination of sparse code and ideal binary system mask | |
CN103400578B (en) | Anti-noise voiceprint recognition device with joint treatment of spectral subtraction and dynamic time warping algorithm | |
CN112820311A (en) | Echo cancellation method and device based on spatial prediction | |
Hamid et al. | Speech enhancement using EMD based adaptive soft-thresholding (EMD-ADT) | |
CN110909827A (en) | Noise reduction method suitable for fan blade sound signals | |
Hassani et al. | Speech enhancement based on spectral subtraction in wavelet domain | |
CN106997766B (en) | Homomorphic filtering speech enhancement method based on broadband noise | |
Yang et al. | A speech enhancement algorithm combining spectral subtraction and wavelet transform | |
CN111009259A (en) | Audio processing method and device | |
CN111968627B (en) | Bone conduction voice enhancement method based on joint dictionary learning and sparse representation | |
CN104064197A (en) | Method for improving speech recognition robustness on basis of dynamic information among speech frames | |
Cao et al. | Study on speech endpoint detection algorithm based on wavelet energy entropy | |
CN110444222A (en) | A kind of speech noise-reduction method based on comentropy weighting | |
CN111402909B (en) | Speech enhancement method based on constant frequency domain transformation |
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 |