Research On Speech Production Model and Phoneme Recognition Using Cooperative Problem Solvers

点击翻译按钮获取中文摘要

1. A vocal tract model was proposed as a speech production model which is possible to be used for phoneme recognition.(1) For building an articulatory model, the 3-dimensional vocal tract shapes of Japanese vowels were measured by magnetic resonance images (MRIs). Vowels were synthesized by using vocal tract area function estimated by MRIs. Analysis of the first and second formants of the synthesized vowels showed good agreement with the subject's original productions.(2) Based on the measurement of the vocal tract shape, midsagittal shape of vocal tract were estimated from 2 positions on frontal tongue surface by a curvature function model of tongue shape, and vocal tract area functions were estimated by the midsagittal shape of vocal tract.2. A cooperative problem solving was used to produce consecutive phoneme based on vocal tract model mentioned above.Cooperative problem solving was used for interactions between the different vocal tract shapes of the different vowels, because there are constraints of the vocal tract shape change from one vowel to another vowel. The system was constructed using PSAs (Problem Solving Agents) developed by the authors. It consists of three kinds of agents : time agent, vowel agent and unity agent. It can produce the pronunciation (e. g. "ya", "wa" and so on) uttered changing vocal tract shape.