|Table of Contents|

Study on State’s Number of HMM of Speech Recognition(PDF)

《南京理工大学学报》(自然科学版)[ISSN:1005-9830/CN:32-1397/N]

Issue:
1998年03期
Page:
16-19
Research Field:
Publishing date:

Info

Title:
Study on State’s Number of HMM of Speech Recognition
Author(s):
Zhang Yan Zhang Jie Huang Zhitong
School of Information, NUST, 210094
Keywords:
speech recog nit ion informat io n theory entr opy hidden Markov mo del
PACS:
TN912.34
DOI:
-
Abstract:
T his paper invest igates state’ s number of HMM ( Hidden Markov Model ) of speech signal on the view of information theory , co nst ruct s a simplified mathemat ical model for studying the number of state of HMM. The resul ts show that the info rmat ion entr opy o f speech sig nal is const ituted of int rinsic ent ropy and addit ional ent ro py . With the number of state increasing , the informat ion ent ropy tends tow ards its intr insic ent ro py . The conclusion is obtained that the number o f state of HMM is proper ly betw een 6 and 8 by considering informat ion ent ro py and calculat ion.

References:

1 Rabiner L R. A tuto rial on hidden M arko v models and selected applicat ions in speech r eco gnition. Pr oc of t he IEEE , 1989, 77( 2) : 257~285
2 周萌清. 信息理论基础. 北京: 北京航空航天大学出版社, 1993
3 马明, 张杰, 王建宇, 黄志同. 语音识别中隐马尔可夫模型初值的估计. 数据采集与处理,1997, 12( 2) : 96~100

Memo

Memo:
-
Last Update: 2013-03-29