Skip to Main content Skip to Navigation

Modélisation du langage pour les systèmes de reconnaissance de la parole destinés aux grands vocabulaires : application à MAUD

Abstract : Automatic speech récognition currently arouses a great interest: it can be considered as a significant field of the man-machine interaction. In fact, we need to communicate with our comput ers using the most natural and direct way: the spoken language; the interaction and the exchange of information are then largely facilitated. At present, several softwares daim to carry out a continuo us speech recognition for a large vocabulary. The performances of these systems are stilliargely lower than those of the human being, particularly on language modeling. The wor k we propose is registered in the framework of language modeling for large vocabularies continuous speech recognition systems. We present in this manuscript new models based on variable length word sequences. These sequences represent linguistic structures or more exactly syntagms. They are automatically detected, on significant text corpora, by using information theory measures. We also propose a hybrid approach combining probabilistic language models, used by the majorîty of current speech recognition systems, with addition al explicit linguistic knowledge. The evaluation of these models is carried out in terms of perplexity and in terms of word prediction with the Shannon game. To test their performances on recognition, we developed a large vocabulary dictation machine named MAUD; it is based on second order hidden Markov models and uses a vocabulary of 20000 words. Compared to the basic version of MA un using a trigram model, the integration of these language models improved the recognition rate by 22% approximately.
Document type :
Complete list of metadata
Contributor : Thèses UL Connect in order to contact the contributor
Submitted on : Thursday, March 29, 2018 - 11:01:47 AM
Last modification on : Saturday, October 16, 2021 - 11:26:08 AM
Long-term archiving on: : Thursday, September 13, 2018 - 7:20:25 PM


Files produced by the author(s)


  • HAL Id : tel-01747502, version 1



Imed Zitouni. Modélisation du langage pour les systèmes de reconnaissance de la parole destinés aux grands vocabulaires : application à MAUD. Autre [cs.OH]. Université Henri Poincaré - Nancy 1, 2000. Français. ⟨NNT : 2000NAN10034⟩. ⟨tel-01747502⟩



Record views


Files downloads