论文部分内容阅读
在人机交互系统中,自动语音识别(ASR)错误将导致交互障碍,通过发起澄清式人机对话可以实现ASR错误恢复。该文提出澄清式人机对话系统结构,用于实现语音识别错误恢复,实现了系统的4个组成部分:ASR错误检测、基于统计机器翻译(SMT)方法的澄清式疑问句生成模型、说话人响应分析、基于有限状态机(FSM)的对话管理模型。各模块均采用与特定任务无关的方法建立。实验结果表明:澄清式人机对话系统可以有效模拟口语中的澄清现象,在不同的错误环境中能够较好的实现ASR错误恢复任务。
In human-computer interaction systems, automatic speech recognition (ASR) errors result in cross-talk and ASR error recovery can be achieved by initiating a clear human-machine dialogue. In this paper, the structure of a clear man-machine dialogue system is proposed to realize the error recovery of speech recognition, and the four components of the system are realized: ASR error detection, clarified interrogative sentence generation model based on statistical machine translation (SMT) method, Analysis, Dialogue Management Model Based on Finite State Machine (FSM). Each module is built using methods that are not task-specific. The experimental results show that the clear man-machine dialogue system can effectively simulate the clarification phenomenon in spoken language and can better achieve ASR error recovery tasks in different error environments.