论文部分内容阅读
基于词网模型的连续语音识别系统在各方面得到了广泛应用,如语音拨号、语音指令、语音菜单、语音导航及语音电话簿等。本文在研究语音识别理论的基础上,设计并开发了基于词网模型的连续语音识别系统—MYASR。MYASR提供了丰富的功能模块,包括前端处理、特征提取、模型训练、词网构建、识别等,使开发一个基于词网模型的连续语音识别应用系统更加方便,同时也是语音识别研究的实验平台。MYASR所采用的XML描述文件,使系统具有良好的可读性和可扩展性。通过在TIMIT语料库上单音子连续语音识别的实验显示,MYASR具有很高的识别性能和实时性能。
The continuous speech recognition system based on word net model has been widely used in all aspects, such as voice dialing, voice commands, voice menu, voice navigation and voice phone book. Based on the research of speech recognition theory, this thesis designs and develops the continuous speech recognition system-MYASR based on word net model. MYASR provides a wealth of functional modules, including front-end processing, feature extraction, model training, word network construction, recognition, etc., so that it is more convenient to develop a continuous speech recognition application system based on the word net model and is also an experimental platform for speech recognition research. The XML description used by MYASR makes the system readable and extensible. Experiments with single tone continuous speech recognition on the TIMIT corpus show that MYASR has high recognition performance and real-time performance.