UNIGE document Chapitre d'actes
previous document  unige:5696  next document
add to browser collection

Comparing Speech Recognizers Derived from Mono- and Multilingual Grammars

Published in Proceedings of TALN/RECITAL. Senlis (France) - june 24-26 2009 - . 2009
Abstract This paper examines the performance of multilingual parameterized grammar rules on speech recognition. We present a performance comparison of two different types of Japanese and English grammar-based speech recognizers. One system is derived from monolingual grammar rules and the other from multilingual parameterized grammar rules. The latter one uses hence the same grammar rules for creation of the language models for these two different languages. We carried out experiments on speech recognition of limited domain dialog application. These experiments show that the language models derived from multilingual parameterized grammar rules (1) perform equally well on both tested languages, on English and Japanese, and (2) that the performance is comparable with the recognizers derived from monolingual grammars that were explicitly developed for these languages. This suggests that the sharing grammar resources between different languages could be one solution for more efficient development of rule-based speech recognizers.
Full text
Proceedings chapter (71 Kb) - public document Free access
(ISO format)
SANTAHOLMA, Marianne Elina. Comparing Speech Recognizers Derived from Mono- and Multilingual Grammars. In: Proceedings of TALN/RECITAL. Senlis (France). [s.l.] : [s.n.], 2009. https://archive-ouverte.unige.ch/unige:5696

302 hits



Deposited on : 2010-04-14

Export document
Format :
Citation style :