本文介绍拼音模型的原理及应用。拼音模型是累加语言模型中同音字的相关数据后得到的3元模型,是在原来的声学模型和语言模型之间增加的一个新环节,可用来求取相关拼音串的先验概率,实验结果表明,用它作为声学层识别的后处理,可使第1名的识别率提高13个百分点,可使前5名的识别率与原来声学模型输出前10 的识别率相当。
Abstract
The principle and application of Pinyin model are introduced in this paper. The Pinyin model is Trigram that adds up same voice data in language model and it is a new link between original acoustic and language model. It can be used to obtain probability of interrelated Pinyin string. The results in experiments show that by using the model to make final process of the recognition of acoustic layer ,the recognition rate of the top one can be increased 13 percent ,and the rate of the front fine is similar with the one of the front ten which are output from original acoustic model.
关键词
声学模型 /
拼音模型 /
语言模型 /
连续语音识别
{{custom_keyword}} /
Key words
acoustic model /
Pinyin model /
language model /
continue speech recognition
{{custom_keyword}} /
{{custom_sec.title}}
{{custom_sec.title}}
{{custom_sec.content}}
参考文献
[1] 黄顺珍,方棣棠. 利用语言模型实现音字转换的高效解码算法. 深圳大学学报,2000 (4)
[2] 郑方,牟晓隆,徐明星等. 汉语语音听写机技术的研究与实现[J] . 软件学报,1999 ,10 (4) :436 - 444
[3] 詹津明,牟晓隆,李树青等. 一个大字表语音识别系统中的语言模型[A] . 第五届全国人机语音通讯学术会议论文集[C] . 哈尔滨:1998 ,214 - 217
[4] 牟晓隆,詹津明,郑方等. 基于修正退化频度估计算法的n - gram语言模型[A] . 第五届全国人机语音通讯学术会议论文集[C] . 哈尔滨:1998 ,206 - 209
[5] Zheng Fang. A syllable-synchronous network search algorithm for word decoding in Chinese speech recognition[A] . IEEE International Conference on Acoustic ,Speech and signal Processing. Phoenix :1999. 601 - 604
[6] Zheng Fang ,Song Zhan-jia ,Xu Ming-xing ,et al. A large-vocabulary speaker-independent Chinese dictation machine[A] . EuroSpeech’99[C] .Budapest ,Hungary :1999 ,2 :819 - 822
[7] 方棣棠,李树青. 汉语语音输入的研究现状与发展前景[A] . 第六届全国人机语音通讯学术会议论文集[C] . 深圳:2001 ,219 - 222
[8] 孙甲松,王作英等. 一个新的汉语连续语音识别系统THEESP[A] . 第五届全国人机语音通讯学术会议论文集[C] . 哈尔滨:1998. 122 - 125
{{custom_fnGroup.title_cn}}
脚注
{{custom_fn.content}}
基金
清华大学智能技术与系统国家重点实验室开发课题基金(00006)
{{custom_fund}}
PDF(124 KB)
