Non-reference voice quality objective assessment method based on deep learning voice enhancement

A speech enhancement and speech quality technology, applied in speech analysis, instruments, etc., can solve the problem of low correlation of subjective evaluation scores, and achieve the effects of high correlation, small root mean square error, and high adaptability

Active Publication Date: 2017-11-17
BEIJING INSTITUTE OF TECHNOLOGYGY
View PDF5 Cites 25 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0008] The purpose of the present invention is to solve the problem that the correlation between the results of the existing non-reference speech objective evaluation algorithm and the subjective evaluat

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
  • Non-reference voice quality objective assessment method based on deep learning voice enhancement
  • Non-reference voice quality objective assessment method based on deep learning voice enhancement
  • Non-reference voice quality objective assessment method based on deep learning voice enhancement

Examples

Experimental program
Comparison scheme
Effect test

Embodiment

[0048] In this example, narrow-band pure speech signals and their distorted signals under 20 kinds of distortion conditions are used as the training data of the speech enhancement model. These distorted voices are scored subjectively by using the MOS method, and used as the training set and test set for the final mapping link.

[0049] The method steps of the present invention are as figure 1 As shown, the details are as follows:

[0050] Step a, the distorted speech signal to be tested is trained based on the speech enhancement model of deep belief network (DBN), obtains the signal after the enhancement, corresponding figure 1 in (1);

[0051] In this step, it is divided into two phases, training phase and enhancement phase. In the training phase, the logarithmic power spectrum information and phase information are first extracted from the pure speech signal and the distorted speech signal, and then the distorted signal parameters are input into the deep belief network. Th...

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 non-reference voice quality objective assessment method based on deep learning voice enhancement, belonging to the voice quality assessment technology field. The non-reference voice quality objective assessment method based on deep learning voice enhancement comprises steps of enabling voice to be tested to go through a trained voice enhancement model based on a deep belief network to obtain an enhanced signal, extracting Mel-frequency cepstral coefficients of signals before and after enhancement respectively, obtaining a difference of the two coefficients, taking the parameter as an input and mapping the parameter as a final objective score through a BP neural network model of a second layer and thus realizing an objective assessment of the non-reference voice quality. Compared with a traditional voice quality assessment model, a correlation degree between the non-reference voice quality objective assessment method of the invention and the object quality score is good and a mean square error is smaller.

Description

technical field [0001] The invention relates to a speech enhancement technology based on a deep belief network and an objective evaluation index mapping technology of an artificial neural network, in particular to a no-reference speech quality objective evaluation method based on a deep learning speech enhancement, and belongs to the field of speech quality evaluation technologies. Background technique [0002] With the development of science and technology, voice communication has become an indispensable part of communication. From telephony to VoIP, voice communication permeates many aspects of our lives. During the transmission of voice communication, the voice compression of the channel, modulation and demodulation, and noise interference will greatly reduce the voice quality, thus reducing the human auditory experience. In order to be able to design a communication system with good transmission performance, judging the performance of the voice communication system has ...

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
IPC IPC(8): G10L25/24G10L25/30G10L25/60
CPCG10L25/24G10L25/30G10L25/60
Inventor 王晶单亚慧孟柳晨谢湘费泽松
Owner BEIJING INSTITUTE OF TECHNOLOGYGY
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Try Eureka
PatSnap group products

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

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap