Speech objective evaluation optimal feature group screening method based on discriminative complementary information

A technology of optimal features and objective evaluation, applied in speech analysis, instruments, etc., can solve the problems of model overfitting and high computational complexity, eliminate the influence of dimension and order of magnitude, improve screening efficiency, and reduce algorithm complexity. Effect

Active Publication Date: 2021-05-07
SOUTH CHINA UNIV OF TECH +2
View PDF6 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0005] The first purpose of the present invention is to overcome the shortcomings and deficiencies of the prior art, and provide a method for screening the optimal feature group of speech objective evaluation based on distinguishing complementary information. This method solves the problem that a single feature is difficult to achieve ideal speech objective evaluation. The combination of features is easy to cause problems of model overfitting and high computational complexity, and can effectively select the best combination of features to achieve objective speech quality evaluation

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
  • Speech objective evaluation optimal feature group screening method based on discriminative complementary information
  • Speech objective evaluation optimal feature group screening method based on discriminative complementary information
  • Speech objective evaluation optimal feature group screening method based on discriminative complementary information

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0075] This embodiment discloses a screening method for the optimal feature group of speech objective evaluation based on discriminative complementary information, which is used to select several kinds of expression features of speech to construct a feature combination that obtains optimal performance, such as figure 1 shown, including the following steps:

[0076] S1. Acquire the voice sample set X={(X n ,s n ),n=1,2,...,N}, each sample X in the voice sample set n Each has a corresponding quality subjective score s n , N is the sample size of the voice sample set, and n is the sample number.

[0077] Then if figure 2 , for each sample X n Extract a variety of candidate features to form a sample feature set:

[0078] S11, carry out filtering preprocessing to sample, adopt voice endpoint detection method (VAD) then to label voiced sound frame, unvoiced sound frame, silent frame in each sample; The zero-rate double-threshold method detects the sample endpoint;

[0079] ...

Embodiment 2

[0116] This embodiment discloses a device for screening optimal feature groups for speech objective evaluation based on discriminative complementary information, which can implement the optimal feature group screening method for speech objective evaluation based on discriminative complementary information described in Embodiment 1, including:

[0117] The sample feature set building block is used to obtain the voice sample set X={(X n ,s n ),n=1,2,...,N}, each sample X in the voice sample set n Each has a corresponding quality subjective score s n , N is the sample size of the voice sample set, n is the sample sequence number, and extracts multiple features to be selected for each sample to form a sample feature set;

[0118] The complementary information entropy calculation module is used to calculate the correlation between the candidate features of each sample, and obtain the complementary information entropy H of the sample feature set R , and the complementary informat...

Embodiment 3

[0127] This embodiment discloses a computer-readable storage medium, which stores a program. When the program is executed by a processor, the method for screening the optimal feature group for objective evaluation of speech based on differentiated complementary information described in Embodiment 1 is implemented. Specifically as follows:

[0128] S1. Acquire the voice sample set X={(X n ,s n ),n=1,2,...,N}, each sample X in the voice sample set n Each has a corresponding quality subjective score s n , N is the sample size of the voice sample set, n is the sample sequence number, and extracts multiple features to be selected for each sample to form a sample feature set;

[0129] S2. Calculate the correlation between the candidate features of each sample, and obtain the complementary information entropy H of the sample feature set R ;

[0130] S3. Calculate the complementary information entropy reduction of the sample feature set in the absence of any single feature, as 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 an optimal feature group screening method for objective speech evaluation based on distinguishing complementary information. The method comprises the following steps of: firstly, extracting various features of a speech sample to form a sample feature set; calculating the complementary information entropy of the feature set and the discrimination of a single feature; according to the correlation between a single feature in the feature set and a subjective score and the distinction of the single feature, selecting a first feature which maximizes the sum of the two features; according to the correlation between the single feature in the candidate feature set and the subjective score and the complementary information entropy of the candidate feature set, selecting other features which maximize the sum of the single feature and the subjective score; and finally, taking a Pearson phase coefficient as a voice objective quality evaluation index, and judging whether the optimal feature group is converged or not according to a performance improvement index of a ridge regression model. According to the method, the problems that ideal objective voice evaluation is difficult to realize by a single feature, model overfitting is easily caused by multi-feature combination, and the calculation complexity is high are solved, and the optimal feature combination for realizing objective voice quality evaluation is effectively selected.

Description

technical field [0001] The invention relates to the technical field of data feature selection, in particular to a screening method for an optimal feature group for objective evaluation of speech based on distinguishing complementary information. Background technique [0002] With the rapid development of communication technology, a variety of codec technologies emerge in an endless stream. Different types of codec technologies and transmission technologies will cause different degrees of damage to voice quality. Because the decline in voice quality will reduce the accuracy of information acquisition, voice The pros and cons of quality directly affect the user experience. In the mobile network environment, if the evaluation of the voice quality of the end user can be performed in real time, the quality can be adjusted according to the result. Therefore, it is very urgent to find an effective, reliable and flexible voice quality evaluation method. [0003] In the literature, ...

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 Applications(China)
IPC IPC(8): G10L25/60G10L19/02
CPCG10L25/60G10L19/02
Inventor 贺前华阳平苏健彬周密陈国强任丹丹李冬梅
Owner SOUTH CHINA UNIV OF TECH
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