Prosody modeling for Uyghur TTS

The prosodic features of syllables such as duration, energy, mean pitch, maximum pitch, minimum pitch and pitch range are extracted from Uyghur TTS speech database and their variations in different prosodic hierarchy are analyzed. Prosodic feature model for Uyghur-language is built by using decision tree and designing contextual features. On that basis, the contextual model based on decision tree is proposed, and prediction performance is compared with the statistical average method. Finally, subjective and objective evaluation is presented.