Dynamic Selection of Acoustic Features in an Automatic Speech Recognition System

Loïc Barrault (1, 2) , Driss Matrouf (2) , Renato de Mori (2) , Roberto Gemello (3) , Franco Mana (3)

1 LIUM - Laboratoire d'Informatique de l'Université du Maine
2 LIA - Laboratoire Informatique d'Avignon
3 LOQUENDO - LOQUENDO

A general approach for integrating different acoustic fea- ture sets and acoustic models is presented. A strategy for using a feature set as a reference and for scheduling the execution of other feature sets is introduced. The strategy is based on the introduction of feature variabil- ity states. Each phoneme of a word hypothesis is as- signed one of such states. The probability that a word hypothesis is incorrect given the sequence of its variabil- ity states is computed and used for deciding the intro- duction of new features. Significant WER reductions have been observed on the test sets of the AURORA3 corpus. Using the CH1 portions of the test sets of the Italian and Spanish cor- pora, word error rate reductions respectively of 16.42% for the Italian and 29.4% for Spanish were observed.

Mots clés: feature combination; automatic speech recognition

Domaines: Traitement du signal et de l'image [eess.SP]; Traitement du signal et de l'image [eess.SP]

Loïc Barrault, Driss Matrouf, Renato de Mori, Roberto Gemello, Franco Mana. Dynamic Selection of Acoustic Features in an Automatic Speech Recognition System. European Signal Processing Conference, EUSIPCO'06, Sep 2006, Italy. pp.N/A. ⟨hal-00433098⟩