EUROSPEECH 2003 - INTERSPEECH 2003
In this paper, we present a statistical approach for phoneme extraction based on universal constraint. Inspired by former phonological studies, we assume a fictitious point in each phoneme that exhibits the most stable information to explain the phoneme's existence. With the universal constraint of phoneme definitions, the point is statistically estimated by an iterative procedure to maximize the local likelihood using a large amount of speech data. We also mention a context dependent modeling of the proposed approach and its integration strategy to obtain more stability. The experimental results show favorable convergencies of both the fictitious points and their likelihoods, which give usefulness for the stable phoneme modeling.
Bibliographic reference. Okawa, Shigeki / Shirai, Katsuhiko (2003): "Statistical estimation of phoneme's most stable point based on universal constraint", In EUROSPEECH-2003, 781-784.