Method for performing pronunciation error detecting based on holding vector machine

A support vector machine and error detection technology, applied in speech analysis, speech recognition, instruments, etc., can solve the problem of sparse manual annotation, achieve good generalization, solve the problem of sparse manual annotation, and achieve good performance.

Inactive Publication Date: 2008-07-30
IFLYTEK CO LTD
View PDF0 Cites 18 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0003] Aiming at the defect of using the posterior probability to detect pronunciation errors in the prior art, the present invention proposes that it can effectively solve the problem of manual labeling sparseness, and make full use of manual labeling of pronunciation error information, thereby ensuring that the error detection model obtained by training can better target different pronunciations. A method of mispronunciation detection using support vector machine for mispronunciation of human, different pronunciation styles

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 for performing pronunciation error detecting based on holding vector machine
  • Method for performing pronunciation error detecting based on holding vector machine
  • Method for performing pronunciation error detecting based on holding vector machine

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0050] Join shown in accompanying drawing 1~6.

[0051] The specific implementation steps of the method for mispronunciation detection using support vector machines are:

[0052] 1. The construction of the speech recognition system, the steps are as follows:

[0053] (1) Collect and train recognizer speech: according to the application needs of language learning, collect or record targeted standard pronunciation corpus in advance, and save them as recognizer training speech files, such as recording standard Mandarin speakers for the Mandarin Chinese proficiency test Putonghua proficiency test corpus;

[0054] (2) Data annotation: Pinyin annotation is performed on the collected standard corpus, so that the collected corpus is targeted for speech evaluation;

[0055] (3) Model training: use HTK to train HMM-based phoneme-level (27 initials, including zero initials, 37 finals) speech recognizer model based on collected standard corpus

[0056] (4) preservation: the model is pr...

Embodiment 2

[0088] The specific implementation steps of the method for mispronunciation detection using support vector machines are:

[0089] 1. The steps to build the speech recognition system are as follows:

[0090] (1) Collect or record standard pronunciation corpus in advance, and save it as a recognizer training voice file;

[0091] (2) Carry out pinyin labeling for the collected standard corpus;

[0092] (3) Model training: train the phoneme-level speech recognizer model according to the collected standard corpus;

[0093] (4) Save the speech recognizer in the library of the computer-aided language learning system.

[0094] 2. Pronunciation error detection feature extraction, the steps are: firstly use the text of the evaluated corpus to segment the pronunciation and calculate the logarithmic likelihood of the target text, denoted as likelihood T , and then, on the boundary obtained by segmentation, calculate the logarithmic likelihood of this segment to all other models in the ...

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 method for detecting mispronunciation based on a support vector machine, which comprises a typical error support vector machine detection method, an abnormal error threshold detection method and an abnormal confusion error support vector machine detection method. The method provided by the invention comprises the following steps: a voice recognition system is established; mispronunciation detection features are extracted; the training objective data of a mispronunciation detection model is acquired; a mispronunciation support vector machine detection model is trained; and mispronunciation is detected; an abnormal mispronunciation detector is trained; a mispronunciation detection threshold is set; wrong pairs of each phoneme which are confused easily are defined; training featured files and target data are generated; a support vector machine model is trained; the mispronunciation detection threshold is set; abnormal confusion errors are determined. The invention can solve the problem of sparse manual marks and ensure that the error detecting model got by means of training can detect errors of different enunciators and different pronunciation styles.

Description

technical field [0001] The invention belongs to the application of automatic speech recognition in pronunciation error detection, and in particular relates to a method for detecting pronunciation errors by using a support vector machine based on automatic speech recognition technology. Background technique [0002] In the method of detecting pronunciation errors based on automatic speech recognition technology, the existing technology mainly relies on the posterior probability, and there are two defects in using the posterior probability as the measurement of pronunciation errors. First, the posterior probability comes from the speech recognizer. Since the artificially labeled error data is too rare, it is difficult to use the posterior probability to update the recognizer model parameters based on the feedback of the pronunciation error labeling data. Therefore, the existing methods are not based on artificial pronunciation errors. The annotation data updates the recognizer...

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): G10L15/00G10L15/06G10L15/02G10L15/10
Inventor 魏思胡国平王海坤刘庆升胡郁刘庆峰吴晓如陈涛陈燕王仁华
Owner IFLYTEK CO LTD
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