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

Method and system for non-parallel corpus-to-speech conversion based on autoregressive network

A speech conversion and autoregressive technology, applied in speech analysis, speech synthesis, speech recognition, etc., can solve problems such as uneven waveform trajectory and pronunciation errors, and achieve the effect of generating smooth waveform trajectory, reducing pronunciation errors, and improving stability.

Active Publication Date: 2022-03-18
中科极限元(杭州)智能科技股份有限公司
View PDF0 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

However, in the actual test process, we found that there is a problem that the waveform track is not smooth, which leads to some mispronunciation

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
  • Method and system for non-parallel corpus-to-speech conversion based on autoregressive network
  • Method and system for non-parallel corpus-to-speech conversion based on autoregressive network
  • Method and system for non-parallel corpus-to-speech conversion based on autoregressive network

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0058] Specific embodiments of the present invention will be described in detail below in conjunction with the accompanying drawings. It should be understood that the specific embodiments described here are only used to illustrate and explain the present invention, and are not intended to limit the present invention.

[0059] Non-parallel corpus speech conversion method based on autoregressive network, obtain phoneme delay probability through pre-trained speech recognition model, use convolutional neural network and gated recurrent unit to model context information in text, and use adaptive attention mechanism Integrating the text features of the current moment and the acoustic features of the previous moment, using the long short-term memory network to predict the acoustic characteristics of the target speaker, and synthesizing speech through the LPCNet vocoder, the naturalness of the converted speech and the similarity of the speaker are improved.

[0060] Such as figure 1 ...

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-parallel corpus speech conversion method and system based on an autoregressive network. The method includes: S1, phoneme delay probability extraction, extracting phoneme delay probability features from input speech; S2, encoding stage, capturing phoneme delay probability The context information in the feature, so as to obtain the text feature representation of the fusion context information; S3, use the adaptive attention mechanism to fuse the text feature at the current moment and the acoustic feature at the previous moment to obtain an augmented feature representation; S4, In the decoding stage, based on the augmented feature representation, the long-short-term memory network is used to predict the acoustic characteristics of the target speaker; S5, speech generation, based on the predicted acoustic characteristics of the target speaker, the speech is synthesized using a vocoder; the system includes: post-phoneme A probabilistic extraction module, an encoding module, a speech generation module, and a set of attention modules and decoding modules.

Description

technical field [0001] The present invention relates to the field of voice conversion, in particular to a method and system for keeping the content of the input voice unchanged but converting the timbre into the target speaker's timbre. Background technique [0002] Speech conversion aims to modify the voice of the original speaker so that the timbre is close to the target speaker, while ensuring that the voice content remains unchanged after conversion. Speech conversion is a very important research topic in the field of artificial intelligence, and has a wide range of applications, such as emotional voice conversion, singing conversion, personalized conversion and so on. [0003] Traditional speech conversion techniques usually require parallel corpus, that is, the original speaker and the target speaker tell the same content. The speech conversion framework based on parallel corpus first uses dynamic time warping technology to obtain the mapping relationship between the ...

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): G10L15/02G10L13/02
CPCG10L15/02G10L13/02
Inventor 连政温正棋
Owner 中科极限元(杭州)智能科技股份有限公司