STARWGAN-GP and x vector-based many-to-many speaker conversion method

A conversion method and speaker technology, applied in neural learning methods, speech analysis, instruments, etc., can solve problems such as GAN training instability and gradient disappearance

Active Publication Date: 2019-04-09
NANJING UNIV OF POSTS & TELECOMM
View PDF7 Cites 14 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0005] Purpose of the invention: The technical problem to be solved by the present invention is to provide a multi-to-many speaker conversion method based on STARWGAN-GP and x-vector, which solves the defect that the speaker's individual characteristics are not fully expressed, and overcomes the existing GAN Problems such as unstable training and gradient disappearance not only further effectively improve the personality similarity of the converted voice, but also improve the quality of the converted voice

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
  • STARWGAN-GP and x vector-based many-to-many speaker conversion method
  • STARWGAN-GP and x vector-based many-to-many speaker conversion method
  • STARWGAN-GP and x vector-based many-to-many speaker conversion method

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0054] Such as figure 1 As shown, the high-quality speech conversion method of the present invention is divided into two parts: the training part is used to obtain the parameters and conversion functions required for speech conversion, and the conversion part is used to convert the source speakers voice into the target speakers voice.

[0055] The implementation steps of the training phase are:

[0056] 1.1) Obtain the training corpus of non-parallel text, the training corpus is the corpus of multiple speakers, including the source speaker and the target speaker. The training corpus is taken from the VCC2018 speech corpus. There are 6 male and 6 female speakers in the training set of this corpus, and each speaker has 81 sentence corpus. This method can realize conversion under parallel text, and can also realize conversion under non-parallel text, so these training corpora can also be non-parallel text.

[0057] 1.2) The training corpus uses the WORLD speech analysis / synthe...

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 STARWGAN-GP and x vector-based many-to-many speaker conversion method which comprises a training stage and a conversion stage. An STARWGAN-GP and an x vector are combined torealize a voice conversion system. According to the method, the X vector with higher characterization performance and practicability is added to characterize personalized features of a speaker, and WGAN-GP is used to replace GAN, so that the problems of unstable GAN training, gradient disappearing and the like are solved; a network with higher training stability and higher convergence speed is constructed, so that the personality similarity and the voice quality of a converted voice are further improved; a high-quality voice conversion method is implemented. The method can relieve the dependency on a parallel text and realize voice conversion under a non-parallel text condition, and also can further integrate a multi-source-target-speaker-pair conversion system into a conversion model torealize many-to-many speaker conversion.

Description

technical field [0001] The invention relates to a many-to-many speaker conversion method, in particular to a many-to-many speaker conversion method based on STARWGAN-GP and x-vector. Background technique [0002] Speech conversion is a research branch in the field of speech signal processing, which is developed and extended on the basis of speech analysis, recognition and synthesis. The goal of voice conversion is to change the voice personality of the source speaker so that it has the voice personality of the target speaker, that is, to make the voice spoken by one person sound like another persons voice after conversion, while preserving semantics . [0003] After years of research on voice conversion technology, many classic conversion methods have emerged. These include most voice conversion methods such as Gaussian Mixed Model (GMM), Recurrent Neural Network (RNN), and Deep Neural Networks (DNN). However, most of these speech conversion methods require that the corpu...

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): G10L13/02G10L15/16G10L15/06G10L25/24G10L25/30G06N3/04G06N3/08
CPCG06N3/08G10L13/02G10L15/063G10L15/16G10L25/24G10L25/30G06N3/045
Inventor 李燕萍曹盼张燕
Owner NANJING UNIV OF POSTS & TELECOMM
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