CN108231084B - Improved wavelet threshold function denoising method based on Teager energy operator - Google Patents
Improved wavelet threshold function denoising method based on Teager energy operator Download PDFInfo
- Publication number
- CN108231084B CN108231084B CN201711260681.5A CN201711260681A CN108231084B CN 108231084 B CN108231084 B CN 108231084B CN 201711260681 A CN201711260681 A CN 201711260681A CN 108231084 B CN108231084 B CN 108231084B
- Authority
- CN
- China
- Prior art keywords
- wavelet
- signal
- threshold
- energy operator
- denoising
- 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
Images
Classifications
-
- G—PHYSICS
- G10—MUSICAL INSTRUMENTS; ACOUSTICS
- G10L—SPEECH ANALYSIS OR SYNTHESIS; SPEECH RECOGNITION; SPEECH OR VOICE PROCESSING; SPEECH OR AUDIO CODING OR DECODING
- G10L21/00—Processing of the speech or voice signal to produce another audible or non-audible signal, e.g. visual or tactile, in order to modify its quality or its intelligibility
- G10L21/02—Speech enhancement, e.g. noise reduction or echo cancellation
- G10L21/0208—Noise filtering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/15—Correlation function computation including computation of convolution operations
Abstract
The invention requests to protect an improved wavelet threshold function denoising method based on a Teager energy operator, and relates to the field of voice signal denoising; by improving the soft and hard threshold denoising function, a new threshold function is provided, and the function not only can overcome the pseudo Gibbs effect caused by the discontinuity of the hard threshold function, but also can overcome the problem of constant deviation after the soft threshold function is denoised. The invention has great improvement on signal-to-noise ratio and mean square error.
Description
Technical Field
The invention belongs to the field of voice signal denoising, and particularly relates to an improved wavelet threshold function denoising method based on a Teager energy operator.
Background
Speech signals often contain noise during acquisition or propagation, wherein the noise seriously affects the subsequent work of signal processing and analysis. Therefore, denoising of speech signals is the most fundamental and important task in the field of signal processing.
The traditional denoising method mainly comprises linear filtering and nonlinear filtering, such as median filtering, Wiener filtering, kalman filtering and the like. The disadvantages of these methods are that it is difficult to reflect the non-stationary characteristics and correlation of the signal while removing noise. In recent years, wavelet transformation is widely applied to signal denoising processing by the characteristic of multi-resolution, can show good signal local characteristics in both time domain and frequency domain, can effectively extract transient information in signals, is suitable for carrying out detailed analysis on non-stationary signals, and obtains better denoising effect. The wavelet threshold denoising is the most used wavelet analysis in denoising application because of simple realization and small calculation amount, and can effectively remove noise and reserve an original voice signal, thereby better improving the signal-to-noise ratio and mean square error of the signal.
The most important in the wavelet threshold denoising process is the selection of the threshold and the threshold function. If the threshold value is too small, the denoising is insufficient, and part of noise is reserved; if the threshold is selected too large, excessive denoising will occur, and weak feature components in the signal will be mistaken for noise and eliminated. The denoising method of the hard threshold and the soft threshold proposed by Donoho and Johnstone is widely applied in practice and achieves better effect, but the discontinuity of the hard threshold function causes the reconstruction signal to easily generate pseudo Gibbs effect; the soft threshold function, although continuous, always has a constant deviation between the estimated value and the actual value. If the threshold value is too small, the denoising is insufficient, and part of noise is reserved; if the threshold is selected too large, excessive denoising occurs, and weak characteristic components in the signal are mistakenly considered as noise to be eliminated, so that part of useful information is lost in the signal. Therefore, in order to obtain better denoising effect, it is important to select an appropriate threshold and a threshold function.
Disclosure of Invention
The present invention is directed to solving the above problems of the prior art. The improved wavelet threshold function denoising method based on the Teager energy operator can improve the signal-to-noise ratio of a noisy signal and reduce the mean square error. The technical scheme of the invention is as follows:
a wavelet threshold function denoising method based on a Teager energy operator is disclosed, and comprises the following steps:
s1, collecting a section of voice signal, and adding noises with different signal-to-noise ratios to the collected voice signal to obtain a voice signal with noises;
s2, carrying out five-layer discrete wavelet decomposition on the noisy speech signal obtained in the step S1 to obtain wavelet decomposition coefficients of each layer;
s3, calculating a Teager energy operator for each layer of wavelet decomposition coefficients obtained in the step S2 to obtain the Teager energy operator value of the wavelet coefficients;
s4, passing the Teager energy operator in the step S3 through a 32-bit Hamming window, then carrying out normalization processing on the value of the Hamming window, and calculating a threshold value in the denoising process;
s5, denoising the noisy speech signal by adopting an improved threshold function according to the threshold in the step S4; the improvement is as follows: the continuity, the progressiveness and the constant deviation of the common threshold function are improved;
and S6, reconstructing the wavelet decomposition coefficient to obtain a denoised voice signal.
Further, step S2 is to perform discrete wavelet decomposition on the noisy speech signal to obtain wavelet coefficients of each layer, where the discrete wavelet decomposition includes:
firstly, defining a wavelet function psi (t), and performing translation and expansion operations on the wavelet function psi (t) to obtain a cluster of wavelet functions psi a, b (t):
a > 0, b ∈ R, a denotes a scale factor b denotes a translation factor, a denotes0And b0All represent the expansion step size, j represents the scale of wavelet decomposition;
discretizing a and b respectively as follows:
a=a0 j,b=ka0 jb0 j,k∈Z,a0≠1
after discretization, a cluster of discrete wavelet functions psi can be obtainedj,k(t):
The discrete wavelet transformed wavelet coefficients of signal f (t) can be expressed as:
Wψrepresents the orthogonal wavelet transform, f represents the speech signal, t represents time, and k represents the number of nodes.
Further, the calculation step of calculating the Teager energy operator in step S3 is as follows:
first, the continuous form of the nonlinear energy operator is defined as:
in the formula (I), the compound is shown in the specification,is the continuous Teager energy operator TEO, x (t) represents a continuous speech signal, when x (n) is a discrete speech signal:
the TEO value is calculated for the wavelet decomposition coefficients: w is aj,m(k) Representing wavelet coefficients T after wavelet decompositionj,m(k) Values representing the Teager energy operators for each layer;
further, the method for calculating the threshold in step S4 is as follows:
smoothing the obtained TEO value, and making it pass through a Hamming window with length of 32 points to obtain M ═ T ═ H, H is Hamming window, T represents Tj,m(k) The shorthand of (1) represents the value of each layer of Teager energy operator; and normalizing M to obtain M':
so that its adaptive threshold can be represented by the expression:
THj,m(k)=λj,m(1-αjM'j,m(k))
in the formula, λj,mRepresenting a threshold, j, m respectively representing the mth subband of the jth layer, αjAre adjustment parameters based on the respective layers.
Wherein N isj,mFor the length of the mth subband at the jth layer, σ j, m represents the standard deviation of gaussian noise:
Further, in step S5, the improved threshold function is:
wherein λ is a threshold value, n is a positive integer, wherein,corresponding to a threshold value which can be automatically adjusted whenj,kWhen | ≧ λ,with | wj,kThe value of l is increasing continuously and the number of the columns,is continuously decreased, and when | wj,kIf λ is less than | is set to 0, a smooth transition region is formed between the noise and the signal.
Further, in step S6, the calculation method for performing wavelet reconstruction on the signal to obtain the denoised signal includes:
where C is a constant independent of the original signal, where Cj,k(t)=<f(t), To indicate psij,k(t)Complex conjugation of (a). The invention has the following advantages and beneficial effects:
firstly, the wavelet coefficient after wavelet decomposition is calculated by Teager energy operator, so that the difference between the noise wavelet coefficient and the signal wavelet coefficient is increased, and the adaptive selection of the threshold value is facilitated; and then, improving a common soft and hard threshold denoising function, and providing a new threshold function, wherein the function not only can overcome the pseudo Gibbs effect caused by the discontinuity of the hard threshold function, but also can overcome the problem of constant deviation after the soft threshold function is denoised.
The invention provides an improved threshold function denoising method based on a Teager energy operator, aiming at the problems of aliasing phenomenon of signal and noise wavelet coefficients, discontinuity of a threshold function at a threshold value, constant deviation of a wavelet coefficient estimated value and an original value and the like in a denoising process. Firstly, calculating a Teager energy operator for a wavelet coefficient, increasing the difference between a signal and noise, and facilitating the selection of a threshold value; and then the wavelet coefficient is quantized by the improved threshold function, so that the problems of pseudo Gibbs effect caused by the discontinuity of the hard threshold function and constant deviation caused by the soft threshold function are effectively avoided. The improved denoising method has the advantages that the denoising effect is greatly improved, the signal-to-noise ratio is greatly improved, the distortion of the voice signal is avoided while denoising is carried out to the maximum degree, and the practicability is high in the actual processing process of the voice signal.
Drawings
FIG. 1 is a flow chart of an improved wavelet threshold function denoising method based on a Teager energy operator according to a preferred embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be described in detail and clearly with reference to the accompanying drawings. The described embodiments are only some of the embodiments of the present invention.
The technical scheme for solving the technical problems is as follows:
as shown in FIG. 1, the invention provides an improved threshold function denoising method based on Teager energy operator, which comprises the following steps:
s1, collecting a section of voice signal by using Cool Edit Pro, and adding noises with different signal-to-noise ratios to the collected voice signal to obtain a voice signal with noises;
s2, calculating Teager energy operators for the wavelet decomposition coefficients of the layers obtained in the step S2 to obtain TEO values of the wavelet decomposition coefficients; (concrete procedure)
The discrete wavelet decomposition comprises the following steps:
first, we define a wavelet function ψ (t), and perform translation and expansion operations on it to obtain a cluster of wavelet functions ψ a, b (t):
we discretize a, b separately as:
a=a0j,b=ka0jb0 j,k∈Z,a0≠1
after discretization, a cluster of discrete wavelet functions psi can be obtainedj,k(t):
The discrete wavelet transformed wavelet coefficients of signal f (t) can be expressed as:
s3, calculating Teager energy operators for the wavelet decomposition coefficients of the layers obtained in the step S2 to obtain TEO values of the wavelet decomposition coefficients;
first, we define the continuous form of the nonlinear energy operator as:
in the formula (I), the compound is shown in the specification,is the continuous Teager energy operator TEO, x (t) represents a continuous speech signal, then when x (n) is a discrete speech signal:
the TEO value is calculated for the wavelet decomposition coefficients:
s4, calculating a threshold value in the denoising process;
smoothing the obtained TEO values, passing them through a hamming window of 32 points in length to obtain M ═ T × H, where × represents the convolution and H is the hamming window, and normalizing M to obtain M':
so that its adaptive threshold can be represented by the expression:
THj,m(k)=λj,m(1-αjM′j,m(k))
wherein j and m respectively represent the m-th sub-band of the j layer, and alphajThe method can reduce the threshold value of the low frequency band and increase the threshold value of the high frequency band based on the adjustment parameters of each layer, and can remove more noise. In the formula (I), the compound is shown in the specification,
wherein the content of the first and second substances,Nj,mis the length of the mth subband of the jth layer. σ j, m represents the standard deviation of gaussian noise:
s5, the step of improving the threshold function is:
for the commonly used threshold functions, soft and hard threshold functions are:
equation (1) represents a hard threshold function, and equation (2) represents a soft threshold function. The voice denoised by the hard threshold function usually has great concussion, and a large amount of noise remains; the wavelet coefficient processed by the soft threshold function has constant deviation with the original wavelet coefficient, so that part of useful high-frequency components are lost, and voice distortion is caused.
In order to overcome the disadvantages of the above two functions, the present invention proposes an improved threshold function:
in the formula, lambda is a threshold value, and n is a positive integer. Wherein the content of the first and second substances,corresponding to a threshold value that can be automatically adjusted. When | wj,kWhen | ≧ λ,with | wj,kThe value of l is increasing continuously and the number of the columns,the constant deviation problem in the soft threshold processing process is avoided by continuously reducing the soft threshold. And when | wj,kWhen the value is less than lambda, the value is not simply set to 0, but a smooth transition region is formed between noise and a signal, and the oscillation possibly generated by direct truncation in a hard threshold value is avoided.
And S6, performing wavelet reconstruction to obtain a denoised voice signal.
The above examples are to be construed as merely illustrative and not limitative of the remainder of the disclosure. After reading the description of the invention, the skilled person can make various changes or modifications to the invention, and these equivalent changes and modifications also fall into the scope of the invention defined by the claims.
Claims (6)
1. An improved wavelet threshold function denoising method based on a Teager energy operator is characterized by comprising the following steps:
s1, collecting a section of voice signal, and adding noises with different signal-to-noise ratios to the collected voice signal to obtain a voice signal with noises;
s2, carrying out five-layer discrete wavelet decomposition on the noisy speech signal obtained in the step S1 to obtain wavelet decomposition coefficients of each layer;
s3, calculating a Teager energy operator for each layer of wavelet decomposition coefficients obtained in the step S2 to obtain the Teager energy operator value of the wavelet coefficients;
s4, passing the Teager energy operator in the step S3 through a 32-bit Hamming window, then carrying out normalization processing on the value of the Hamming window, and calculating a threshold value in the denoising process;
s5, denoising the noisy speech signal by adopting an improved threshold function according to the threshold in the step S4; the improvement is as follows: the continuity, the progressiveness and the constant deviation of the common threshold function are improved; in step S5, the improved threshold function is:
wherein λ is a threshold value, n is a positive integer, wherein,corresponding to a threshold value which can be automatically adjusted whenj,kWhen | ≧ λ,with | wj,kThe value of l is increasing continuously and the number of the columns,is continuously decreased, and when | wj,kWhen the value is less than lambda, a smooth transition region is formed between noise and a signal instead of 0;
and S6, performing wavelet reconstruction on the signal subjected to the denoising processing of S5 to obtain a denoised voice signal.
2. The improved wavelet threshold function denoising method based on Teager' S energy operator as claimed in claim 1, wherein said step S2 performs discrete wavelet decomposition on the noisy speech signal to obtain wavelet coefficients of each layer, the discrete wavelet decomposition step is:
firstly, defining wavelet function psi (t), and making it undergo the processes of translation and expansion operation to obtain a cluster of wavelet functions psia,b(t):
a denotes a scale factor b denotes a translation factor, a0And b0All represent an extension stepLength, j, represents the scale of the wavelet decomposition;
discretizing a and b respectively as follows:
a=a0 j,b=ka0 jb0 j,k∈Z,a0≠1
after discretization, a cluster of discrete wavelet functions psi can be obtainedj,k(t):
The discrete wavelet transformed wavelet coefficients of signal f (t) can be expressed as:
3. The method for denoising the wavelet threshold function based on the Teager energy operator as claimed in claim 2, wherein the step of calculating the Teager energy operator in the step S3 is:
first, the continuous form of the nonlinear energy operator is defined as:
in the formula (I), the compound is shown in the specification,is the continuous Teager energy operator TEO, x (t) represents a continuous speech signal, when x (n) is a discrete speech signal:
the TEO value is calculated for the wavelet decomposition coefficients: w is aj,m(k) Representing wavelet coefficients T after wavelet decompositionj,m(k) Values representing the Teager energy operators for each layer;
4. the improved wavelet threshold function denoising method based on Teager energy operator as claimed in claim 3, wherein the threshold value calculating method in step S4 is:
smoothing the obtained TEO value, and making it pass through a Hamming window with length of 32 points to obtain M ═ T ═ H, H is Hamming window, T represents Tj,m(k) The shorthand of (1) represents the value of each layer of Teager energy operator; and normalizing M to obtain M':
so that its adaptive threshold can be represented by the expression:
THj,m(k)=λj,m(1-αjM'j,m(k))
in the formula, λj,mRepresenting a threshold, j, m respectively representing the mth subband of the jth layer, αjAre adjustment parameters based on the respective layers.
6. The Teager energy operator-based improved wavelet threshold function denoising method of claim 1, wherein in step S6, the calculation method for performing wavelet reconstruction on the signal to obtain the denoised signal comprises:
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711260681.5A CN108231084B (en) | 2017-12-04 | 2017-12-04 | Improved wavelet threshold function denoising method based on Teager energy operator |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711260681.5A CN108231084B (en) | 2017-12-04 | 2017-12-04 | Improved wavelet threshold function denoising method based on Teager energy operator |
Publications (2)
Publication Number | Publication Date |
---|---|
CN108231084A CN108231084A (en) | 2018-06-29 |
CN108231084B true CN108231084B (en) | 2021-09-10 |
Family
ID=62653137
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201711260681.5A Active CN108231084B (en) | 2017-12-04 | 2017-12-04 | Improved wavelet threshold function denoising method based on Teager energy operator |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108231084B (en) |
Families Citing this family (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109611703B (en) * | 2018-10-19 | 2021-06-22 | 宁波鄞州竹创信息科技有限公司 | LED lamp convenient to installation |
CN109697702A (en) * | 2018-11-02 | 2019-04-30 | 浙江工业大学 | Medical ultrasound image denoising method based on bending wave conversion |
CN110808059A (en) * | 2019-10-10 | 2020-02-18 | 天津大学 | Speech noise reduction method based on spectral subtraction and wavelet transform |
CN113541729B (en) * | 2021-07-12 | 2022-06-07 | 电子科技大学 | Time-frequency graph denoising method based on time-frequency matrix |
CN114091983B (en) * | 2022-01-21 | 2022-05-10 | 网思科技股份有限公司 | Intelligent management system for engineering vehicle |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2003081578A1 (en) * | 2002-03-21 | 2003-10-02 | U.S. Army Medical Research And Materiel Command | Methods and systems for detecting, measuring, and monitoring stress in speech |
CN101625869A (en) * | 2009-08-11 | 2010-01-13 | 中国人民解放军第四军医大学 | Non-air conduction speech enhancement method based on wavelet-packet energy |
CN103575807A (en) * | 2013-10-24 | 2014-02-12 | 河海大学 | Method for detecting structural damage of beam based on Teager energy operator-wavelet transformation curvature mode |
CN106030707A (en) * | 2014-02-14 | 2016-10-12 | 唐纳德·詹姆士·德里克 | System for audio analysis and perception enhancement |
CN107274908A (en) * | 2017-06-13 | 2017-10-20 | 南京邮电大学 | Small echo speech de-noising method based on new threshold function table |
-
2017
- 2017-12-04 CN CN201711260681.5A patent/CN108231084B/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
WO2003081578A1 (en) * | 2002-03-21 | 2003-10-02 | U.S. Army Medical Research And Materiel Command | Methods and systems for detecting, measuring, and monitoring stress in speech |
CN101625869A (en) * | 2009-08-11 | 2010-01-13 | 中国人民解放军第四军医大学 | Non-air conduction speech enhancement method based on wavelet-packet energy |
CN103575807A (en) * | 2013-10-24 | 2014-02-12 | 河海大学 | Method for detecting structural damage of beam based on Teager energy operator-wavelet transformation curvature mode |
CN106030707A (en) * | 2014-02-14 | 2016-10-12 | 唐纳德·詹姆士·德里克 | System for audio analysis and perception enhancement |
CN107274908A (en) * | 2017-06-13 | 2017-10-20 | 南京邮电大学 | Small echo speech de-noising method based on new threshold function table |
Non-Patent Citations (2)
Title |
---|
"A semisoft thresholding method based on Teager energy operation on wavelet packet coefficients for enhancing noisy speech";Tahsina Farah Sanam等;《EURASIP Journal on Audio, Speech, andMusic Processing》;20131231;全文 * |
"基于Teager能量算子的自适应小波语音增强";高亚召 等;《语音技术》;20091231;第33卷(第01期);全文 * |
Also Published As
Publication number | Publication date |
---|---|
CN108231084A (en) | 2018-06-29 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108231084B (en) | Improved wavelet threshold function denoising method based on Teager energy operator | |
CN107274908B (en) | Wavelet voice denoising method based on new threshold function | |
CN109643554B (en) | Adaptive voice enhancement method and electronic equipment | |
Soon et al. | Speech enhancement using 2-D Fourier transform | |
Rui et al. | Online wavelet denoising via a moving window | |
CN110808059A (en) | Speech noise reduction method based on spectral subtraction and wavelet transform | |
US10083705B2 (en) | Discrimination and attenuation of pre echoes in a digital audio signal | |
Fu et al. | Perceptual wavelet adaptive denoising of speech. | |
CN105869652A (en) | Psychological acoustic model calculation method and device | |
Sun et al. | An adaptive speech endpoint detection method in low SNR environments | |
CN102509268B (en) | Immune-clonal-selection-based nonsubsampled contourlet domain image denoising method | |
CN115293219A (en) | Wavelet and kurtosis fused pulse signal denoising method | |
Sun et al. | Speech enhancement via two-stage dual tree complex wavelet packet transform with a speech presence probability estimator | |
CN106997766B (en) | Homomorphic filtering speech enhancement method based on broadband noise | |
Surendran et al. | Perceptual subspace speech enhancement with variance normalization | |
Liu | A new wavelet threshold denoising algorithm in speech recognition | |
Wang et al. | Extraction of weak crack signals based on sparse code shrinkage combined with wavelet packet filtering | |
Verma et al. | A comparative performance analysis of wavelets in denoising of speech signals | |
CN114822577B (en) | Method and device for estimating fundamental frequency of voice signal | |
Sanam et al. | A DCT-based noisy speech enhancement method using teager energy operator | |
Zhang et al. | Histogram equalization and noise masking for robust speech recognition | |
Zhao et al. | A robust algorithm for formant frequency extraction of noisy speech | |
Liu et al. | A Weighting Threshold Optimization Method in Speech Recognition | |
Senthamizh Selvi | Speech Enhancement using Adaptive Filtering with Different Window Functions and Overlapping Sizes | |
Salavedra et al. | Comparison of different order cumulants in a speech enhancement system by adaptive Wiener filtering |
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 |