Applying pitch connection control in Mandarin speech synthesis

In this paper, a novel tone-based pitch connection control in unit selection is described to improve naturalness of output speech for Mandarin text-to-speech (TTS) baseline system. This study mainly focuses on pitch connections of concatenative syllables. To improve the concatenation quality, we apply offset pitch of preceding syllable and onset pitch of following syllable in unit selection. According to the statistical result on corpus, three types of pitch connection constraints are proposed. Based on the property of pitch connection constraint, corresponding tone-based cost functions play important role in unit selection for continuity improving at concatenation point. By applying the defined cost functions in unit selection, more suitable units are selected and more natural-sounding synthesized speech is achieved.