[1]赵力,邹采荣,吴镇扬.语音识别中基于MCE准则的CHMM学习方法[J].东南大学学报(自然科学版),2000,30(3):21-25.[doi:10.3969/j.issn.1001-0505.2000.03.005]
 Zhao Li,Zou Cairong,Wu Zhenyang.Minimum Recognition Error Training of CHMM for Speech Recognition[J].Journal of Southeast University (Natural Science Edition),2000,30(3):21-25.[doi:10.3969/j.issn.1001-0505.2000.03.005]
点击复制

语音识别中基于MCE准则的CHMM学习方法()
分享到:

《东南大学学报(自然科学版)》[ISSN:1001-0505/CN:32-1178/N]

卷:
30
期数:
2000年第3期
页码:
21-25
栏目:
计算机科学与工程
出版日期:
2000-05-20

文章信息/Info

Title:
Minimum Recognition Error Training of CHMM for Speech Recognition
作者:
赵力 邹采荣 吴镇扬
东南大学无线电工程系,南京 210096
Author(s):
Zhao Li Zou Cairong Wu Zhenyang
Department of Radio Engineering,Southeast University, Nanjing 210096
关键词:
MCE CHMM 语音识别
Keywords:
MCE CHMM speech recognition
分类号:
TP391.42;TN912.3
DOI:
10.3969/j.issn.1001-0505.2000.03.005
摘要:
研究了用于CHMM的MCE学习方法.提出了简单形式的识别函数、误差函数、损失函数等的定义方法以及总损失函数的最小化方法.通过对非特定人汉语连续语音识别实验,证明了提出的函数形式和损失函数最小化方法对CHMM的MCE学习的有效性.
Abstract:
This paper describes an implementation of the minimum recognition error training for continuous Gaussian density HMM. We defined the misrecognition measure directly by discriminant function for the minimum recognition error training of CHMM. Also a minimizing method of the discriminant error function for CHMM training was investigated. The experiment on speaker-independent continuous speech recognition of Chinese shown that the proposed method is effective.

参考文献/References:

[1] Chou W,Lee C H,Juang B H.Minimum error rate training based on n-best string models.Proc ICASSP93,1993,2:653~655
[2] Juang B H,Katagiri S.Discriminative learning for minimum error classification.IEEE Trans Signal Processing,1993,40(12):3043~3054
[3] Rainton D,Sagayama S.Minimun error classification training of DHMM.J Acoust Soc Jpn,1992,E-13(6):333~339
[4] Zhao L,Kobayashi Y,Niimi Y.Recognition of chinese continuous speech based on the integration of phonetic and prosodic information.ICCCPOL’97,1997:178~183

备注/Memo

备注/Memo:
基金项目:国家自然科学资金资助项目(69871009).
第一作者:男, 1958年生, 博士后.
更新日期/Last Update: 2000-05-20