The adaptation of machine conversational speed to speaker utterance speed in human-machine communication

Abstract

The development of a machine that adapts to speech of any human speaker and responds at an optimal utterance speed is considered. The relationship between speech speed and speech temporal parameters (mean silence duration S), mean talkspurt duration T, mean talkspurt rate (1/T(T+S)), and speech activity (T/(T+S)) in the case of human conversation at normal speed, determined from 46 subjects, is discussed. This was investigated in Japanese language for the range of 'hangover' and 'fill-in' values used to bridge short silence durations. Based on the above relationship, a method for adapting machine conversational speed to speaker utterance speed is proposed. The method controls the silence durations in the machine's active speech so that the machine's speech activity equals the speaker's speech activity, with a fill-in value of 130 ms. The effectiveness of this method is demonstrated on the basis of the degree of adaptation to speech speed. >