Chinese speech recognition method based on pinyin constraint joint learning

A technology of speech recognition and Chinese character recognition, which is applied in speech recognition, neural learning methods, speech analysis, etc., can solve problems such as difficult convergence of Chinese character recognition, and achieve the effect of improving the recognition effect

Pending Publication Date: 2021-02-09
KUNMING UNIV OF SCI & TECH
View PDF1 Cites 1 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0003] The invention provides a Chinese speech recognition method based on joint learning of pinyin constraints, which is used to introduce pinyin as a constraint on Chinese character decoding in Chinese speech recognition, which can promote the model to learn better phonetic features, and alleviate the current system's difficulty in recognizing Chinese characters convergence problem

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
  • Chinese speech recognition method based on pinyin constraint joint learning
  • Chinese speech recognition method based on pinyin constraint joint learning
  • Chinese speech recognition method based on pinyin constraint joint learning

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0040] Embodiment 1: as figure 1 As shown, the Chinese speech recognition method based on the joint learning of pinyin constraints, the specific steps of the Chinese speech recognition method based on the joint learning of pinyin constraints are as follows:

[0041] Step1. Collect pinyin texts corresponding to speech and Chinese character texts; on the public training corpus data_aishell, collect pinyin texts corresponding to speech and Chinese character texts, so as to obtain speech, Chinese text, pinyin text training sets, test sets and verification sets ;

[0042] Step2, shared encoder; the shared encoder uses a 4-layer convolutional network and a 5-layer bidirectional LSTM. The bidirectional LSTM has 512 hidden state units in each direction. During model training, it can perceive the supervision signals of pinyin and Chinese characters at the same time , thus introducing an inductive bias closer to Chinese phonetics.

[0043] Step3, pinyin speech recognition; in the deco...

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 relates to a Chinese speech recognition method based on pinyin constraint joint learning, and belongs to the technical field of natural language processing. According to the method, firstly, pinyin texts corresponding to voices and texts are collected from a public Chinese corpus set, secondly, speech features are encoded through a shared encoder, then pinyin speech recognition is used as an auxiliary task, and then pinyin is used as a decoding constraint in the decoding process. Pinyin speech recognition and Chinese speech recognition are combined for learning based on a sharedencoder, inductive bias closer to speech is introduced, and the expression ability of the encoder for Chinese speech is enhanced. According to the Chinese speech recognition method based on pinyin constraint joint learning provided by the invention, the word error rate of Chinese recognition is reduced, and powerful support is provided for subsequent work such as pinyin fusion and pinyin error correction in the Chinese speech recognition process; and the problem that the end-to-end model is difficult to converge in Chinese character recognition is relieved.

Description

technical field [0001] The invention relates to a Chinese speech recognition method based on joint learning of pinyin constraints, belonging to the technical field of natural language processing Background technique [0002] In the field of automatic speech recognition, the current speech recognition model has achieved good results in phonograms such as English and French. However, Chinese is a typical ideographic script, and there is no direct correspondence between Chinese characters and phonetics. However, Pinyin, as a symbol for the pronunciation of Chinese characters, has an internal connection with Chinese characters. The cascade method of recognizing phonetic features as syllable (pinyin) units and converting pinyin to Chinese characters through a conversion model has error propagation. In order to avoid this problem, the Chinese character-pinyin recognition model uses pinyin to help the recognition of Chinese characters during training. , but the recognition effect ...

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): G10L15/26G10L15/187G06N3/08G06N3/04
CPCG10L15/187G06N3/049G06N3/084
Inventor 余正涛梁仁凤王振晗朱俊国高盛祥毛存礼
Owner KUNMING UNIV OF SCI & 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