単語音声の音素系列による表現

概要

論文の詳細を見る
We are studying a system of speech recognition which is composed of acoustic and linguistic analysis parts. This report concerns with the acoustic part, where we are trying to express the speech wave by a sequence of phonemes. We have introduced a psychological model which consists of three fields, say, sensory, psychological and decision. In the sensory field variations of some characteristic features of speech waves are perceived and under the influence of the state of psychological field. The psychological field is excited mainly by spectral informations and selects phoneme candidates. Decision of stationarity of the speech wave is done in the sensory field using a linear discriminant function of stationarity. Final decision of phonemes is done in decision field which is made up of the outputs from both sensory and psychological fields, and the state of psychological field is represented as the final output whenever the sensory field is stationary. We have succeeded to express each 50 words uttered by 3 males pretty well by phoneme strings containing vowels, nasals, stops and / s, r, y, w /. Our model is working in the speech recognition system for statements in Fortran language.
山梨大学の論文