Looking for breakthrough ideas for innovation challenges? Try Patsnap Eureka!

mmse-lsa speech enhancement method based on improved noise estimation

A MMSE-LSA, speech enhancement technology, applied in speech analysis, instruments, etc., can solve problems such as noise residue, achieve the effects of reducing speech distortion, suppressing noise, and reducing errors

Active Publication Date: 2021-09-07
JINGWAH INFORMATION TECH CO LTD
View PDF3 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0006] The purpose of the present invention is to overcome the shortcomings of the prior art that the noise estimation error cannot be reduced and the speech signal-to-noise ratio can be increased while avoiding noise residue and speech distortion in a non-stationary environment, and a MMSE based on improved noise estimation is provided. -LSA speech enhancement method, through the construction of the entropy ratio of speech feature parameters, using the smoothed speech existence probability to accurately track noise changes, reduce noise residue and speech distortion, and improve speech quality

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • mmse-lsa speech enhancement method based on improved noise estimation
  • mmse-lsa speech enhancement method based on improved noise estimation
  • mmse-lsa speech enhancement method based on improved noise estimation

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0071] like figure 1 As shown, the present embodiment relates to a MMSE-LSA speech enhancement method based on improved noise estimation, comprising the following steps:

[0072] S1: Framing and windowing the noisy speech, and then doing short-time Fourier transform to obtain the amplitude spectrum and phase angle of the noisy speech;

[0073] S2: Calculate the logarithmic energy and spectral entropy of the noisy speech according to the result of step S1, and construct a new speech characteristic parameter energy-entropy ratio;

[0074] S3: According to the properties of the energy entropy ratio and the voice existence probability in step S2, the energy entropy ratio and the voice existence probability are proportional to each other, and the mathematical relationship model between the energy entropy ratio and the voice existence probability is established to obtain the estimated value of the voice existence probability;

[0075] S4: Smooth the estimated value of the speech ex...

Embodiment 2

[0085] like figure 1 As shown, the present embodiment is based on Embodiment 1. In the step S2, the logarithmic energy is usually significantly larger than the non-speech segment according to the short-term energy of the speech segment, specifically as follows,

[0086] If it is assumed that the noisy speech signal of the i-th frame after frame division and windowing is , then the short-term energy of the frame is:

[0087]

[0088] Among them, N is the frame length, further improving the energy calculation to obtain the logarithmic energy:

[0089]

[0090] In the formula, α takes 2.1.

Embodiment 3

[0092] like figure 1 As shown, on the basis of Embodiment 1 or 2 in this embodiment, in the step S2, the spectral entropy can be obtained by the following formula,

[0093] Let the speech signal of the i-th frame after windowing and framing of the noisy speech signal be , after Fourier transform, let the power spectrum of the kth frequency component be , then the normalized probability density function of each frequency component is:

[0094]

[0095] Then the spectral entropy of each analysis frame is:

[0096] .

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

PUM

No PUM Login to View More

Abstract

The invention discloses a MMSE-LSA speech enhancement method based on improved noise estimation, comprising: S1: performing frame division and windowing processing on the noisy speech, performing short-time Fourier transform, and obtaining the amplitude spectrum of the noisy speech and phase angle; S2: Calculate the logarithmic energy and spectral entropy of noisy speech, and construct a new speech feature parameter energy entropy ratio; S3: Obtain the proportional relationship between the energy entropy ratio and the probability of speech existence, and establish the energy entropy ratio and speech The mathematical relationship model of the existence probability is used to obtain the estimated value of the speech existence probability; S4: Smooth the estimated value of the speech existence probability, and use the smoothed speech existence probability to update the noise power spectrum estimation; S5: Calculate the prior signal-to-noise ratio to obtain Spectral gain estimation, adding a constraint threshold to the gain function; S6: Use MMSE-LSA spectral estimator to perform speech enhancement on noisy speech; through the construction of speech feature parameter energy entropy ratio, reduce noise residual and speech distortion, and achieve improvement Voice quality purposes.

Description

technical field [0001] The invention relates to the field of noise-improved speech enhancement, in particular to an MMSE-LSA speech enhancement method based on improved noise estimation. Background technique [0002] Noise pollution will obliterate useful information in speech, seriously affecting speech quality and intelligibility. Speech enhancement technology is a technology that suppresses noise and improves voice quality while minimizing distortion. The current speech enhancement methods are mainly based on short-term spectrum amplitude estimation method, speech parameter model method, auditory scene analysis method and so on. [0003] The MMSE-LSA method is based on voice activity detection. It uses the prior knowledge of the statistical characteristics of voice and noise to judge the noise frame and the voice frame for the noisy voice frame. The noise estimate is only updated in the noise frame, and the voice frame is continued. In the past noise spectrum estimation...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

Application Information

Patent Timeline
no application Login to View More
Patent Type & Authority Patents(China)
IPC IPC(8): G10L21/0208G10L21/0216G10L25/21G10L25/45
CPCG10L21/0208G10L21/0216G10L25/21G10L25/45
Inventor 冯谦
Owner JINGWAH INFORMATION TECH CO LTD
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Patsnap Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Patsnap Eureka Blog
Learn More
PatSnap group products

Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic, Popular Technical Reports.

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap|About US| Contact US: help@patsnap.com