使用期待提高对话系统的语音识别率

在回顾了各种语言模型的基础上,针对如何更有效地构建口语对话系统中语音识别器的语言模型展开讨论,研究并实现了使用系统期待来建立语言模型的方法.在口语对话系统中,根据系统提出的问题或者系统给用户的提示,对话管理器产生对用户响应的期待,也称作系统期待.由于系统期待是建立在对话系统当前状态的基础上,所以可根据系统当前状态构建系统期待,从而建立更加优化的语言模型,并使用此语言模型来提高语音识别的识别率.

Improving the recognition rate of speech for dialogue system using expectation

Based on all kinds of the language models for speech recognizer, it is discussed how to make language model more efficient in speech recognizer. The method of building the language model based on the expectation of system is realized. In the spoken dialog system, dialog manager may product the expectation of the user response called expectation of system according to the system questions or system prompts. Since the expectation of system is based on the current state of system, the expectation of system can be constructed according to the system state. From which, a better language model for improving speech recognition rate can be constructed.