We research on speech recognition for utterance in a scarcely audible voice. We think that when a person utters, even if it is without audible sound, he produces muscle activities similar to the activities made with audible sound. Our purpose is to recognize utterance made without audible sound based on surface EMG signals. In this paper, we select 4 circumoral places to measure EMGs in order to recognize the 5 Japanese vowel sounds. In our analysis of measured EMG waveforms, we discovered that the waveforms contain clues to distinguish vowel sounds and the features of utterance such as strength of voice.
展开▼