从大规模语料库中选择不定长度的拼接单元的合成方法.pdfVIP

SELECTING NON-UNIFORM UNITS FROM A VERY LARGE CORPUS FOR CONCATENATIVE SPEECH SYNTHESIZER Min Chu, Hu Peng, Hong-yun Yang, Eric Chang Microsoft Research China Beijing, P.R.C Email: minchu@ ABSTRACT and how to obtain the best representative instances for each unit are still issues that haven’t been studied well. This paper proposes a two-module TTS structure, which bypasses the prosody model that predicts numerical prosodic This paper investigates the above issues based on an ultimate parameters for synthetic speech. Instead, many instances of each assumption that we have a very large speech corpus that contains basic unit from a large speech corpus are classified into enough prosodic and spectral varieties for all synthetic units. categories by a CART, in which the expectation of the weighted This assumption is valid under a constraint that the whole corpus sum of square regression error of prosodic features is used as retains the same speaking style, which is referred as the “relax splitting criterion. Better prosody is achieved by keeping slender reading style”, the same speech rate and the same timbre. Since diversity in prosodic features of instances belong to the same no pitch or duration modification will be applied to the selected class. A multi