Proceedings of the Xth Conference, Helsinki 2008
Edited By Martti Vainio, Reijo Aulanko and Olli Aaltonen
24. Prosodic model for Russian 249
PROSODIC MODEL FOR RUSSIAN Nina B.Volskaya Pavel A. Skrelin 1 Introduction A new prosodic model for Russian TTS is ideologically and technically different from the previously developed rule-based Russian concatenation TTS. Since prosodic information is extracted from natural speech, it uses a selection of natural contours from recorded speech material. The aim of any TTS model is to produce natural, or as close to natural speech as possible. Quality and naturalness of the synthesized speech to a great extent depend on the units used for concatenation. For segments we use allophones. In this prosody model we have tried to follow the same principle and include phonetic realizations of selected phonological intonation patterns. 2 Material The experimental material was recorded for the project “High Quality Russian Speech Synthesis System” (State contract = 02.524.11.4001). The corpus included words for extracting allophones to be used in the allophone- and sub- allophone based speech synthesis, and recorded texts — for extracting prosodic information for Unit Selection technology. Recordings were made from 8 native Russian professional (4 male and 4 female) speakers aged 35–40. They all read the same texts, representing different speech styles: 2 narrative stories which included parts of conversation, 2 news- paper articles (on politics and on the IT-technology), and a play. The recording session took place in a sound proofed room at the Experimental Phonetics Lab of Saint-Petersburg State University. The total recording time was over 6 hours. The article deals with part of the analyzed material: recordings of 2 narrative texts...
You are not authenticated to view the full text of this chapter or article.
This site requires a subscription or purchase to access the full text of books or journals.
Do you have any questions? Contact us.Or login to access all content.