Human-machine dialogue clarification system for speech recognition error recovery

Incorrect automatic speech recognition(ASR) result hinder interactions in human-machine systems.This problem can be solved by clarifying the dialogue for ASR error recovery.This paper presents a human-machine dialogue clarification system that includes ASR error detection,clarification question generation based on statistical machine translation(SMT),user response analysis,and dialogue clarification management based on a finite state machine(FSM).All metrics are not task specific.Tests show that the system can effectively clarify misunderstandings by handling mis-recognized utterances to achieve high accuracy error recovery.