Effect of HMM Parameter on Robustness of Chinese Isolated Words Recognition

DOI

10.2991/icmt-13.2013.48How to use a DOI?

Keywords

Isolated Words Recognition, Number of HMM States, Number of HMM Observation Symbols, Speech Recognition Robustness

Abstract

The characteristics of speech signals are stable in short-term but unstable in long-term, so speech sequence modeling method based on Markov Chain can more effectively represent the feature of speech signals. According to some basic modeling unit, this method can also constructs sentence model of continuous speech, and the accuracy and flexibility of this method are relatively high. This paper firstly constructs a Chinese isolated word speech system whose vocabulary is small, and made the recognition rate achieve 100%. Secondly, this paper changes the number of HMM (Hidden Markov Model) states and the number of HMM observation symbols to test the effect that these two parameters have on the recognition robustness. The experimental results show that increasing the number of HMM states and the number of observation symbols can improve the robustness of isolated word speech recognition. When the values of these two parameters are greater than some certain value, continuing to increase these two parameters has no obvious effect on the recognition robustness.

Copyright

© 2013, the Authors. Published by Atlantis Press.

Open Access

This is an open access article distributed under the CC BY-NC license (http://creativecommons.org/licenses/by-nc/4.0/).

Download article (PDF)

TY  - CONF
AU  - De Liu
AU  - Mingjiang Wang
AU  - Zejun Wu
PY  - 2013/11
DA  - 2013/11
TI  - Effect of HMM Parameter on Robustness of Chinese Isolated Words Recognition
BT  - Proceedings of 3rd International Conference on Multimedia Technology(ICMT-13)
PB  - Atlantis Press
SP  - 390
EP  - 400
SN  - 1951-6851
UR  - https://doi.org/10.2991/icmt-13.2013.48
DO  - 10.2991/icmt-13.2013.48
ID  - Liu2013/11
ER  -