Reconnaissance automatique de phonèmes guidée par les syllabes
Olivier Le Blouch, Patrice Collen.
This paper presents a phonetic transcription system of
French speech. This recognizer is based on a phonetic
transcription driven by syllables, a syllabic bigram
language modelling and a HMM topology adapted to
syllables. The phone error rate obtained is compared to
basic, usual systems at phonetic level : once the
resulting syllables have been converted to phones, the
phone error rate on a 12 minute-part of BREF80 corpus
is as low as 15.8% with 35 phones.