120 集中系モデルを用いた音声生成シミュレーション(音響・構造のモデル化・最適化技術)

Abstract

Larynx cancer or pulmonary problems cause initial symptoms such as hoarse voice. There is a possibility that making diagnoses of state of affected area by analyzing these change of voice. In this study, our purpose is to enable diagnoses of larynx cancer or pulmonary problems by an acoustic analysis model inside of the vocal tract. As a first step, we aimed to establish an acoustic analysis method in the vocal tract and identification method of vocal tract shapes and vocal sound source waveforms. In this paper, We modeled the air in the vocal tract by a concentrated mass model consists of masses, connecting linear springs, connecting dampers, and base support dampers, and made an acoustic analysis model in the vocal tract. Then we estimated vocal tract shapes by using Levenberg-Marquardt method, and analyzed vocal sound source waveforms by using transfer matrix method from measured voice wave. And we simulated a speech production by the analyzed sound source. The identified vocal tract shapes are valid and the analyzed vocal sound source waveforms are similar to the Rosenberg wave. The speech production results are clear voices. Then, we conclude our proposed methods are valid.