Generalization of Thai tone contour in HMM-based speech synthesis

This paper presents an improvement of vowel tone modeling from using an idea of generalization of tone contour by sharing F0 probability stream in the conventional HMM-Based Speech Synthesis system. As of the characteristic of parameter-based speech synthesis, the excitation source (F0) is highly significant to the correctness of synthetic speech especially for a tonal language like Thai. This paper is proposing a method of sharing F0 probability distributions within the same syllabic tones by tying them together before adjusting their probabilities. The implementation and evaluation was done on HTS platform which is based-on HTK. The results show that the proposed method can increase Thai tone correctness at about 14.1% objectively and 90% subjectively.