The Research On Speech Rate In Speech Synthesis

Nowadays the speech synthesis technology is becoming more and more muture and the speech synthesis systems could synthesize speech with high intelligibility. However, there are still large gaps between these machine-generated utterances and human natural speech, such as speech rate. Under the current speech synthesis systems, the controlling strategy of the speech rate and effect is not very well, so the synthesized speech is uncomfortable and even wearisome. As a part of the project, which is cooperated by the Phonetics Lab of Institute of Linguistics, Chinese Academy of Social Sciences and Motorola China Research Center, this paper tries to analyse the prosodic features affected by different rates of English utterances and Chinese discourse, paving the way for getting a better speech rate strategy.At first, speech rate effects on prosodic features of English utterances are investigated. Speech rate effects on prosodic features are statistically analyzed for fast, normal and slow speech, including prosodic structure variation, prosodic duration variation, F0 distribution and variation, and accent placement variation.Secondly, the controlling patterns of speech rate on Chinese Mandarin discourse are investigated. We also analyse the prosodic features such as F0 distribution, accent palacement variation under three speech rates. Moreover, we inspect the controlling strategy of speech rate and find that the interplay between speech rate and prosodic features is non-linear.Finally, a synthetic algorithm idea is proposed based on the above research.