CN102945674A - Method for realizing noise reduction processing on speech signal by using digital noise reduction algorithm - Google Patents
Method for realizing noise reduction processing on speech signal by using digital noise reduction algorithm Download PDFInfo
- Publication number
- CN102945674A CN102945674A CN2012105051890A CN201210505189A CN102945674A CN 102945674 A CN102945674 A CN 102945674A CN 2012105051890 A CN2012105051890 A CN 2012105051890A CN 201210505189 A CN201210505189 A CN 201210505189A CN 102945674 A CN102945674 A CN 102945674A
- Authority
- CN
- China
- Prior art keywords
- signal
- noise
- algorithm
- subspace
- 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.)
- Pending
Links
Images
Landscapes
- Soundproofing, Sound Blocking, And Sound Damping (AREA)
- Filters That Use Time-Delay Elements (AREA)
Abstract
The invention relates to a method for realizing noise reduction processing on a speech signal by using a digital noise reduction algorithm. The method comprises the following steps of: firstly, processing a signal with noise by using a subspace method so as to obtain a relatively pure processed signal; and subsequently inputting a self-adaptive filter to eliminate speech noise by using an LMS (least mean square) algorithm. According to the method, a subspace noise reduction algorithm and the self-adaptive LMS algorithm are combined and utilized so as to filter the noise; and the subspace can suppress the noise relevance well but can not work well in an area with a high signal to noise ratio, and the defect of the subspace can be made up by using the self-adaptive LMS algorithm, so that a good noise reduction effect can be achieved with the combination of the subspace noise reduction algorithm and the self-adaptive LMS algorithm.
Description
Technical field
The present invention relates to a kind of noise processing method, particularly a kind of digital noise reduction algorithm that utilizes is realized denoise processing method to voice signal.
Background technology
Along with the development of digital signal processing theory and application electric technology and perfect, the speech electronic noise reduction system has obtained increasing application in people's life, extracting purified signal under the strong noise background all is the problem of a hot topic in research and application.Build filtering circuit from early stage employing analog device and solve noise problem to modern times employing dsp program loading noise reduction algorithm, reduced the realization difficulty, improved the realization effect.
The voice de-noising algorithm also grows out of nothing, and the spectral subtraction that mainly contains the Steven.F.Boll proposition of profound influence and the language noise reduction algorithm of the opportunity Wiener filtering that Lim proposes are wherein arranged the most.Modern study shows that a certain algorithm of simple usefulness comes not obvious to the noisy speech treatment effect of full frequency band and do not have universality.
Summary of the invention
The present invention be directed to a certain algorithm of simple usefulness and come the unconspicuous problem of noisy speech treatment effect to full frequency band, proposed a kind of digital noise reduction algorithm that utilizes and realized denoise processing method to voice signal, subspace noise reduction algorithm and ADAPTIVE LMS ALGORITHM are united to be made noise is carried out filtering.The subspace can well suppress Noise Correlation, but can not fine work in the high s/n ratio zone; ADAPTIVE LMS ALGORITHM can remedy the shortcoming of subspace on this, and both are in conjunction with reaching good noise reduction.
Technical scheme of the present invention is: a kind of digital noise reduction algorithm that utilizes is realized comprising the steps: denoise processing method to voice signal
The first step: after the voice signal process A/D analog to digital conversion of collection, output noisy speech signal x (n) processes it with subspace method first, obtains relatively pure processing signals:
Wherein subspace method is processed each frame voice is carried out following six steps processing:
1) the covariance R of calculating noise signal
n, use formula
Estimated matrix
, R
nTo estimate by each frame independent sample voice;
2) right
Carry out Eigenvalues Decomposition:
, the eigenvalue matrix=eigenvectors matrix of estimated matrix;
3) basis
Eigenwert estimate the space size of voice signal subspace;
4) the μ value formula by the subspace calculates the μ value,
, μ wherein
0=4.2, s=6.25, SNR
Db=10log
10SNR, SNR are signal to noise ratio (S/N ratio);
5) linear estimated value calculates by following formula:
6) pass through
Estimate the voice signal of enhancing, x (n) is noisy speech signal, and p (n) is through the output after the algorithm process of subspace;
Second step, the voice signal input adaptive wave filter that strengthens is eliminated voice noise with least mean square algorithm, through output reducing noise of voice signal after the D/A digital-to-analog conversion.
Square error in the described second step in the least mean square algorithm
Be expressed as:
Wherein s (n) is clean speech signal n sample value constantly,
Be signal s (n) linear predictor value, N (n) is noise signal, and
Beneficial effect of the present invention is: the present invention utilizes the digital noise reduction algorithm to realize denoise processing method to voice signal, and the relative merits of zygote space noise reduction algorithm and ADAPTIVE LMS ALGORITHM are carried out combination and complementation, reach obvious noise reduction.
Description of drawings
Fig. 1 is that the present invention utilizes the digital noise reduction algorithm to realize the entire block diagram that the denoise processing method of voice signal is realized;
Fig. 2 is that the present invention utilizes the digital noise reduction algorithm to realize LMS in the denoise processing method of voice signal is processed block diagram.
Embodiment
Because we can't obtain the covariance of clean speech signal, so we only have from following expressions of noise and obtain estimated matrix
:
(1）
R
nBe the covariance matrix of noise signal,
R y The covariance matrix of purified signal,
IIt is noise component.A topmost estimation that the factor is the μ value when using the subspace noise reduction algorithm.Excessive μ value estimate meeting so that during estimating background noise comprising cost excessive, opposite too small μ value can increase noise remnants.Our purpose is to reduce voice distortion at speech space, strengthens the noise remove amount at spatial noise, and the noise that this paper adopts is estimated based on quick SNR(Signal-to-Noise Ratio signal to noise ratio (S/N ratio)) formula:
Wherein
, μ
0With s be a constant.Can well finish the estimation to the μ value, but have a potential problem, that is exactly in the situation of high SNR, and the μ value of estimating according to this formula is not best.
Given this, we need to adopt LMS algorithm (Least mean square algorithm, i.e. least mean square algorithm) to come this result who estimates lower computing is compensated.The LMS algorithm also has the estimation of a μ value, and it is that filter joint exponent number and input signal power are estimated, in order to compensate the leak after the algorithm process of subspace, we estimate numerical value in conjunction with two kinds, adopt following computing formula to unify the μ value of LMS sum of subspace:
μ wherein
0=4.2, s=6.25, SNR
Db=10log
10SNR.
The entire block diagram that realizes as shown in Figure 1 will be carried out by the voice signal that the microphone collection is come the A/D analog-to-digital conversion module and be processed, and then realize the simulating reality noise signal is carried out noise reduction with noise reduction algorithm.Among the figure: x (n) is noisy speech signal; P (n) is through the output after the algorithm process of subspace; D (n) is p (n) time delayed signal.
Whole processing procedure mainly was divided into for two steps, and the latter's input is as the former output:
A, with subspace method noisy speech signal x (n) is processed first, obtain relatively pure processing signals;
B, utilize the LMS algorithm to strengthen the problem that subspace method is left under the high ground unrest.
For realizing desired content, next analyze above-mentioned A, the specific implementation method in two steps of B:
The steps A subspace method is processed (time-domain constraints), each frame voice is carried out following six steps process:
1) the covariance R of calculating noise signal
n, with formula (1) estimated matrix
, R
nTo estimate by each frame independent sample voice;
2) right
Carry out Eigenvalues Decomposition:
Gauche form is respectively eigenvalue matrix and the eigenvectors matrix of estimated matrix;
3) suppose
Eigenwert by descending sort, estimate the space size of voice signal subspace with following formula:
4) calculate the μ value by formula (3);
5) linear estimated value calculates by following formula:
6) pass through
Estimate the voice signal of enhancing, Signal estimation=optimum solution * is with the speech signal of making an uproar.
The LMS algorithm is strengthened among the step B:
At first, adaptive LMS noise processed mechanism: LMS processes block diagram as shown in Figure 2.Sef-adapting filter has widely to be used, and comes voice noise is eliminated with the LMS algorithm on this basis, and p among the figure (n) is through the output after the algorithm process of subspace;
The error output signal
(5),
P (n) is through the output after the algorithm process of subspace, and y (n) is every output through a LMS processing.
Square error
Be expressed as:
Suppose that the A/D sampling period is T, establish s (n) and be clean speech signal n sample value constantly, then with s (n) contiguous L in the past moment sample value can be expressed as:
A wherein
n=[a
1, a
2..., a
L], S
N-1=[s (n-1), s (n-2) ... s (n-L)], a
nBe current coefficient.
Generally voice are polluted by the neighbourhood noise N of additivity (n), and the adaptive process of LMS algorithm is exactly automatically to regulate the weights of wave filter, so that square error
Reach minimum value (
The estimated value of N (x)).Because noise is less at different moment coefficient of autocorrelation from the uncorrelated and noise of voice signal, so can use (6) formula to estimate.
Claims (2)
1. one kind is utilized the digital noise reduction algorithm to realize it is characterized in that denoise processing method to voice signal, comprises the steps:
The first step: after the voice signal process A/D analog to digital conversion of collection, output noisy speech signal x (n) processes it with subspace method first, obtains relatively pure processing signals:
Wherein subspace method is processed each frame voice is carried out following six steps processing:
1) the covariance R of calculating noise signal
n, use formula
Estimated matrix
, R
nTo estimate by each frame independent sample voice;
2) right
Carry out Eigenvalues Decomposition:
, the eigenvalue matrix=eigenvectors matrix of estimated matrix;
3) basis
Eigenwert estimate the space size of voice signal subspace;
4) the μ value formula by the subspace calculates the μ value,
5) linear estimated value calculates by following formula:
6) pass through
Estimate the voice signal of enhancing, x (n) is noisy speech signal, and p (n) is through the output after the algorithm process of subspace;
Second step: the voice signal input adaptive wave filter that strengthens is eliminated voice noise with least mean square algorithm, through output reducing noise of voice signal after the D/A digital-to-analog conversion.
2. the described digital noise reduction algorithm that utilizes is realized denoise processing method to voice signal it is characterized in that the square error in the described second step in the least mean square algorithm according to claim 1
Be expressed as:
，
Wherein s (n) is clean speech signal n sample value constantly,
Be signal s (n) linear predictor value, N (n) is noise signal, and
, a
nBe coefficient.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2012105051890A CN102945674A (en) | 2012-12-03 | 2012-12-03 | Method for realizing noise reduction processing on speech signal by using digital noise reduction algorithm |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2012105051890A CN102945674A (en) | 2012-12-03 | 2012-12-03 | Method for realizing noise reduction processing on speech signal by using digital noise reduction algorithm |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN102945674A true CN102945674A (en) | 2013-02-27 |
Family
ID=47728612
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN2012105051890A Pending CN102945674A (en) | 2012-12-03 | 2012-12-03 | Method for realizing noise reduction processing on speech signal by using digital noise reduction algorithm |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN102945674A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105744456A (en) * | 2016-02-01 | 2016-07-06 | 沈阳工业大学 | Digital hearing-aid self-adaptive sound feedback elimination method |
| CN105874535A (en) * | 2014-01-15 | 2016-08-17 | 宇龙计算机通信科技(深圳)有限公司 | Speech processing method and speech processing apparatus |
| CN112671358A (en) * | 2020-11-27 | 2021-04-16 | 天津城建大学 | Design method of comprehensive signal generation processing system |
Citations (4)
| 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 |
| US6408269B1 (en) * | 1999-03-03 | 2002-06-18 | Industrial Technology Research Institute | Frame-based subband Kalman filtering method and apparatus for speech enhancement |
| CN101778322A (en) * | 2009-12-07 | 2010-07-14 | 中国科学院自动化研究所 | Microphone array postfiltering sound enhancement method based on multi-models and hearing characteristic |
| CN101894561A (en) * | 2010-07-01 | 2010-11-24 | 西北工业大学 | Wavelet transform and variable-step least mean square algorithm-based voice denoising method |
-
2012
- 2012-12-03 CN CN2012105051890A patent/CN102945674A/en active Pending
Patent Citations (4)
| 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 |
| US6408269B1 (en) * | 1999-03-03 | 2002-06-18 | Industrial Technology Research Institute | Frame-based subband Kalman filtering method and apparatus for speech enhancement |
| CN101778322A (en) * | 2009-12-07 | 2010-07-14 | 中国科学院自动化研究所 | Microphone array postfiltering sound enhancement method based on multi-models and hearing characteristic |
| CN101894561A (en) * | 2010-07-01 | 2010-11-24 | 西北工业大学 | Wavelet transform and variable-step least mean square algorithm-based voice denoising method |
Non-Patent Citations (4)
| Title |
|---|
| YI HU等: "A Generalized Subspace Approach for Enhancing Speech Corrupted by Colored Noise", 《IEEE TRANSACTIONS ON SPEECH AND AUDIO PROCESSING》, vol. 11, no. 4, 31 July 2003 (2003-07-31), pages 334 - 341 * |
| 吴怡: "基于LMS算法的语音增强系统的研究", 《中国优秀硕士学位论文全文数据库》, 31 October 2011 (2011-10-31) * |
| 徐望等: "一种基于信号子空间和听觉掩蔽效应的语音增强方法", 《信号处理》, vol. 20, no. 2, 30 April 2004 (2004-04-30) * |
| 章旭景: "基于卡尔曼滤波的语音增强算法研究", 《中国优秀硕士学位论文全文数据库》, 31 July 2010 (2010-07-31), pages 136 - 92 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105874535A (en) * | 2014-01-15 | 2016-08-17 | 宇龙计算机通信科技(深圳)有限公司 | Speech processing method and speech processing apparatus |
| CN105744456A (en) * | 2016-02-01 | 2016-07-06 | 沈阳工业大学 | Digital hearing-aid self-adaptive sound feedback elimination method |
| CN112671358A (en) * | 2020-11-27 | 2021-04-16 | 天津城建大学 | Design method of comprehensive signal generation processing system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105741849B (en) | The sound enhancement method of phase estimation and human hearing characteristic is merged in digital deaf-aid | |
| CN108831495B (en) | Speech enhancement method applied to speech recognition in noise environment | |
| TWI466107B (en) | Multi-microphone robust noise suppression | |
| US8175871B2 (en) | Apparatus and method of noise and echo reduction in multiple microphone audio systems | |
| CN105679330B (en) | Based on the digital deaf-aid noise-reduction method for improving subband signal-to-noise ratio (SNR) estimation | |
| CN101901602B (en) | Method for reducing noise by using hearing threshold of impaired hearing | |
| CN105390142B (en) | A kind of digital deaf-aid voice noise removing method | |
| CN101083640A (en) | Low complexity noise reduction method | |
| CN104810024A (en) | Double-path microphone speech noise reduction treatment method and system | |
| JP2003534570A (en) | How to suppress noise in adaptive beamformers | |
| CN109767782A (en) | A kind of sound enhancement method improving DNN model generalization performance | |
| CN102982808A (en) | Speech denoising device and method based on wavelet transform | |
| Yang et al. | A noise reduction method based on LMS adaptive filter of audio signals | |
| CN107180643A (en) | One kind is uttered long and high-pitched sounds sound detection and elimination system | |
| CN102945674A (en) | Method for realizing noise reduction processing on speech signal by using digital noise reduction algorithm | |
| CN107731242B (en) | Gain function speech enhancement method for generalized maximum posterior spectral amplitude estimation | |
| CN103578466A (en) | Voice and non-voice detection method based on fractional order Fourier transformation | |
| CN107680610A (en) | A kind of speech-enhancement system and method | |
| CN106328160B (en) | Noise reduction method based on double microphones | |
| CN105869649B (en) | Perceptual filtering method and perceptual filter | |
| CN113409804A (en) | Multichannel frequency domain speech enhancement algorithm based on variable-span generalized subspace | |
| CN112652322A (en) | Voice signal enhancement method | |
| CN106060717A (en) | High-definition dynamic noise-reduction pickup | |
| TW526468B (en) | System and method for eliminating background noise of voice signal | |
| CN110602621B (en) | Noise reduction method and system for digital hearing aid and special DSP |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C12 | Rejection of a patent application after its publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20130227 |