Looking for breakthrough ideas for innovation challenges? Try Patsnap Eureka!

Small-scale corpus DNN-HMM acoustic model

An acoustic model, small-scale technology, used in speech analysis, speech recognition, instruments, etc., can solve problems such as uneven data distribution, small scale of labeled data, and models that cannot describe speech features well.

Pending Publication Date: 2018-12-21
INNER MONGOLIA UNIV OF TECH
View PDF4 Cites 7 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0007] The purpose of the present invention is to provide a small-scale corpus DNN-HMM acoustic model to solve the problem of a large number of initial problems in the process of training the DNN-HMM acoustic model under the small-scale corpus proposed in the background technology due to the small scale of labeled data and unbalanced data distribution. The parameters have not been updated, and the model cannot describe the speech features in the corpus well, resulting in a decline in the recognition rate

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
  • Small-scale corpus DNN-HMM acoustic model
  • Small-scale corpus DNN-HMM acoustic model
  • Small-scale corpus DNN-HMM acoustic model

Examples

Experimental program
Comparison scheme
Effect test

Embodiment approach

[0076] A small-scale corpus DNN-HMM acoustic model, characterized in that: in the small-scale corpus speech recognition of the DNN-HMM acoustic model, feature extraction is first performed on the input small-scale corpus speech, and the extracted features are used to analyze the DNN-HMM The acoustic model is trained, and the DNN-HMM acoustic model is obtained; the language model is trained by using the text information corresponding to the speech of the small-scale corpus, and the language model of the small-scale corpus is obtained; the decoder is obtained by using the acoustic model, the language model and the dictionary to obtain The entire small-scale corpus speech recognition framework;

[0077] The entire small-scale corpus speech recognition process includes two stages of training and recognition:

[0078] The training phase includes speech database and feature extraction, DNN-HMM acoustic model, text database, language model, dictionary, speech decoding and search algo...

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 a small-scale corpus DNN-HMM acoustic model. For small-scale corpus speech recognition in a DNN-HMM acoustic model, feature extraction is carried out on an inputted small-scalecorpus speech; the DNN-HMM acoustic model is trained by using the extracted feature and the DNN-HMM acoustic model is obtained; a language model is trained by using text information corresponding to the small-scale corpus speech to obtain a small-scale corpus language model; and a decoder is obtained by using the acoustic model, the language model and dictionary construction, so that an overall small-scale corpus speech recognition frame is obtained.

Description

technical field [0001] The present invention relates to the technical field of acoustics, and more specifically, to a small-scale corpus DNN-HMM acoustic model. Background technique [0002] Since the deep neural network has the characteristics of automatically extracting data features and memorizing data features according to data characteristics in modeling, and at the same time, it does not make any assumptions about the distribution of data, so it is widely used in machine learning. [0003] Therefore, deep neural networks are introduced in acoustic modeling for speech recognition. [0004] However, when modeling a deep neural network, a large amount of corpus data is required to saturate the neural network for training, so that the modeling effect can be better and meet the needs of practical applications. [0005] In the process of training the DNN-HMM acoustic model on a small-scale corpus, due to the small scale of labeled data and the uneven distribution of data, a...

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): G10L15/06G10L15/14G10L15/16
CPCG10L15/06G10L15/063G10L15/144G10L15/16
Inventor 马志强陈艳李图雅
Owner INNER MONGOLIA UNIV OF TECH
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Patsnap Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Patsnap Eureka Blog
Learn More
PatSnap group products

Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic, Popular Technical Reports.

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap|About US| Contact US: help@patsnap.com