Unlock instant, AI-driven research and patent intelligence for your innovation.

Many-to-many voice conversion method based on beta-vae

A voice conversion, many-to-many technology, applied in voice analysis, instruments, etc., can solve the problems of insufficient representation ability and difficult expansion, achieve high-quality voice conversion, improve voice quality, and improve the effect of insufficient representation

Active Publication Date: 2021-09-07
NANJING UNIV OF POSTS & TELECOMM
View PDF5 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0004] Purpose of the invention: the technical problem to be solved by the present invention is to provide a beta-VAE-based many-to-many voice conversion method, which solves the problem that the hidden variables in the existing VAE network have insufficient characterization capabilities for voice data and are difficult to expand to more complex Insufficient voice data, better improve the voice quality after conversion, and effectively improve the conversion performance

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
  • Many-to-many voice conversion method based on beta-vae
  • Many-to-many voice conversion method based on beta-vae
  • Many-to-many voice conversion method based on beta-vae

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0031] Such as figure 1 As shown, the present embodiment provides a many-to-many voice conversion method based on beta-VAE under a non-parallel text condition, which is divided into two steps of training and conversion:

[0032] 1. Speaker voice training stage

[0033] 1.1 Obtain non-parallel training corpus. The speech library used here is VCC2018, which contains 8 source speakers (SF1, SF2, SM1, SM2, SF3, SF4, SM3, SM4) and 4 target speakers (TF1 , TF2, TM1, TM2). The non-parallel training corpus selected in this paper is 4 source speakers: SF3, SF4, SM3, SM4, and 4 target speakers TF1, TF2, TM1, TM2. Among them, S (source) represents the source speaker, T (target) represents the target speaker, F (female) represents female, and M (male) represents male. Since the goal of this paper is non-parallel speech conversion, the selected training corpus is also non-parallel. There are 4 source speakers: SF3, SF4, SM3, SM4, and 4 target speakers TF1, TF2, TM1, TM2. The content of...

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 many-to-many speech conversion method based on beta‑VAE. By introducing adjustable parameters β and C, the modification of the variational autoencoder (VAE) framework is completed, while improving the latent variable disentanglement ability. , and also improve the problem of insufficient coding ability at the bottleneck layer, and realize multi-speaker-to-multi-speaker voice conversion. This method solves the problem that hidden variables in the existing VAE network have insufficient representation ability for voice data and it is difficult to expand to more complex voice data, and can better improve the converted voice quality and effectively improve the conversion performance. Moreover, this method removes the dependence on parallel texts, the training process does not require any alignment operation, and can also integrate the conversion systems of multiple source-target speaker pairs into one conversion model, that is, realize many-to-many conversion.

Description

technical field [0001] The invention relates to a many-to-many voice conversion method, in particular to a beta-VAE-based many-to-many voice conversion method. Background technique [0002] After years of research on speech conversion technology, many classic conversion methods have emerged, including Gaussian Mixed Model (GMM), frequency bending, deep neural network (Deep Neural Network, DNN) and methods based on unit selection Wait. However, most of these speech conversion methods need to use parallel corpora for training, that is, the source speaker and the target speaker need to utter sentences with the same speech content and speech duration, and the pronunciation rhythm and emotion should be as consistent as possible. However, in the practical application of speech conversion, it is not easy to obtain a large amount of parallel corpus, or even unsatisfactory. In addition, the accuracy of the alignment of speech feature parameters during training has also become a cons...

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): G10L21/007G10L21/013
CPCG10L21/007G10L21/013
Inventor 李燕萍张成飞许吉良张燕
Owner NANJING UNIV OF POSTS & TELECOMM