Speech Prosody 2002
There is a large amount of segmental variants in a natural speech corpus. It is very important to label those variants correctly for a corpus based TTS system. We successfully applied automatic triphone segmentation to a large speech corpus with syllable segmentation and prosodic annotation. In this paper, we also report (1) recognition error analysis based on prosodic structure, and (2) the relationship between coarticulation phenomena and prosodic position.
Bibliographic reference. Zu, Yiqing / Zheng, Hong (2002): "Effect of prosodic structure on segmental variants", In SP-2002, 739-742.