Estimating vocal tract shapes of Thai vowels from contextual vowel variation

Abstract

This paper presents a computational estimation of vocal tract shape parameters as articulatory targets of Thai vowels in an articulatory synthesizer, by means of analysis-by-synthesis with acoustic data as input. A speech corpus designed to capture the contextual variants of nine Thai long vowels, consisting of 81 disyllabic utterances, was recorded by a native Thai speaker. For each utterance, two targets, one for each syllable, were estimated by optimizing the target parameters to minimize the MFCC error between original and synthesized speech. An analysis-by-synthesis approach was used to iteratively optimize the shape parameters. The estimated targets of each vowel type were then averaged, resulting in nine articulatory targets, each corresponding to a vowel. The optimized targets were then used to synthesize Thai vowels both in monosyllables and in disyllabic sequences. The results indicate that the estimated targets effectively represent the underlying articulatory goals of Thai vowels.