清华大学电子工程系,北京,100084
纸质出版:2002
移动端阅览
单翼翔, 张昊天, 李虎生, 等. 邮包校核语音识别系统的实时实现[J]. 电子学报, 2002,30(4):544-547.
SHAN Yi-xiang, ZHANG Hao-tian, LI Hu-sheng, et al. The Implementation of a Practical Chinese Speech Recognition System For The Parcel Post Checking Task[J]. Acta Electronica Sinica, 2002, 30(4): 544-547.
本文研究开发了一套邮包信息校核语音识别系统.该系统利用中大词汇量非特定人连续语音识别技术实时实现了邮包信息的语音校核.系统可以识别普通话或四川话语音
可识别的词汇量约为4500条.系统还采用了拒识技术与说话人自适应技术
提高了整个系统的稳健性.实验表明对普通话的首选识别率达到98.7%
前三选识别率达到99.9%.对四川话的首选识别率达到95.9%
前三选识别率达到98.6%
对无关语音的正确拒识率达到85%
对口音较重的说话人经过自适应后识别率可提高5-8个百分点.
In this paper a real-time Parcel Post Checking Speech Recognition System is developed.This system can achieve the parcel post information checking and verification by means of a medium vocabulary speaker-independent continuous speech technology.Input speech can be Mandarin or Sichuan dialect.Vocabulary size is about 4500.By employing the advantages of the rejection model and speaker adaptation technology
the system's robustness is improved remarkably.The experiments show that the system has a very good performance.For mandarin voice input
the recognition accuracy of the top one candidate is 98.7%
and the recognition accuracy of the top three candidates is 99.9%; For Sichuan dialect one the recognition accuracy of the top one candidate is 95.9%
and the recognition accuracy of the top three candidates is 98.6%.
0
浏览量
954
下载量
1
CSCD
关联资源
相关文章
相关作者
相关机构
京公网安备11010802024621