TCN-Transform-CTC-based end-to-end Chinese speech recognition method

A technology of speech recognition and Chinese, applied in speech recognition, speech analysis, instruments, etc., can solve problems such as little research on Mandarin ASR tasks, achieve the effect of facilitating the overall performance and improving the recognition rate

Active Publication Date: 2022-02-08
匀熵科技无锡有限公司
View PDF5 Cites 11 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0004] However, the work cited above basically focuses on English ASR tasks. English speech recognition generally constructs models through pronunciation units such as subwords, CI phonemes, context-dependent phonemes, and word models, and has achieved good results. Attention-based Mandarin ASR tasks are rarely studied

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
  • TCN-Transform-CTC-based end-to-end Chinese speech recognition method
  • TCN-Transform-CTC-based end-to-end Chinese speech recognition method
  • TCN-Transform-CTC-based end-to-end Chinese speech recognition method

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0028] 1. Connect temporal classification and attention mechanism

[0029] 1.1 Connectionist temporal classifi-cation (CTC)

[0030] The CTC model was developed by Graves et al [Graves A, Fernández S, Gomez F, et al. Connectionist temporal classification: labeling unsegmented sequence data with recurrent neural networks[C] / / Proceedings of the 23rd international conference on Machine learning.2006:369-376.] A time series classification method is proposed. Compared with traditional speech recognition, which needs to be pre-aligned, it directly maps the input audio sequence to a system of words or other modeling units (such as phonemes and characters), which greatly simplifies the speech recognition model. built and trained. At the same time, CTC introduces a blank label, so that the network can be buffered when judging the current input speech frame, and solves the problem of aligning repeated characters and continuous labels.

[0031] First, CTC introduces a latent variable C...

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 provides an end-to-end Chinese speech recognition method based on TCN-Transform-CTC (Chinese speech recognition-Transform-CTC), and belongs to the field of speech recognition. In order to solve the problems in the prior art, firstly, a time sequence convolutional neural network (TCN) is used for enhancing capture of position information by a neural network model, and secondly, connection time sequence classification (CTC) is fused on the basis, so that a TCN-Transform-CTC model which is better in recognition effect and higher in generalization is provided. Under the condition of not using any language model, the experimental result on the open source speech database AISHELL-1 of the Hilshell Chinese mandarin shows that the word error rate of TCN-Transform-CTC is relatively reduced by 10.91% compared with that of Transform, and the final word error rate of the model is 5.31%.

Description

technical field [0001] The invention belongs to the field of speech recognition, and in particular relates to a TCN-Transformer-CTC end-to-end Chinese speech recognition method. Background technique [0002] Automatic Speech Recognition (ASR) technology can enable people to communicate more smoothly between people and machines. At present, with the rapid development of speech recognition technology, speech recognition technology is widely used in smart customer service, smart furniture, vehicle systems, robots, etc. Wide range of applications. The traditional continuous speech recognition system is composed of multiple complex modules, including training the acoustic model based on Hidden Markov Model (HMM), constructing pronunciation dictionary and language model, so it is a complicated project. The general steps are as follows: first, a pronunciation dictionary designed by professional linguists is needed, then the phoneme sequence generated by the acoustic model is mappe...

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): G10L15/16G10L19/16
CPCG10L15/16G10L19/16
Inventor 孙俊
Owner 匀熵科技无锡有限公司
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