Bibliografické údaje¶
Název | Reduction of Computing Cost in Unit Selection TTS System with Large Corpus |
Autor | Kala, J., Matoušek, J., Tihelka, D. |
Typ publikace | Článek ve sborníku |
Sborník | Proceedings of the 19th Czech-German Workshop on Speech Processing |
Místo konání | Prague, Czech Republic |
Strana | 85-91 |
Rok | 2009 |
ISBN | 978-80-86269-18-4 |
Abstrakt¶
Unit selection concatenative speech synthesis requires high amount of computingtime, especially for very large corpus. The corpus for our Czech text-to-speechsystem ARTIC contains over 650 thousands of diphone tokens. Unit selectionsystem with such a large corpus is not well suitable for run-time speech generationwhen standard Viterbi search is used. In case of wide-range dialogue systems orother server applications, it is required to dispatch many requests at one moment andthis could be problematic in relation to server resources. This paper describes variousmodifications of Viterbi search algorithm for massive reduction of computation costin unit selection approach to speech synthesis.