Recognition of Words With The Help of The Seraciroise Expert System

Auteurs-es

  • xavier
  • Martine Gérard
  • Guy Mercier

Résumé

In order to test the performance of the acoustic-phonetic decoding module on the Seraciroise expert system, we have implemented a lexical analyzer, the function of which is to match each word of the task vocabulary against the phonetic hypotheses lattice. A one-stage dynamic comparison algorithm, initially designed for global recognition of connected words, has been adapted. Our knowledge-based approach makes it possible to improve performance significantly with the help of heuristics, e.g. concerning local constraints and the measure of similarity. Introducing phonological, syllable and prosodic information into the lexicon allows refinement of the strategy by basing on islands of reliability, Such phonological phenomena as merging, spreading, insertion, deletion and confusion are dealt with in a rather flexible way: likelihood weights, penalty factors and thresholds of reliability are determined according to the most encountered recognition errors. The object - and rule-based representation gives advanced opportunities for system extension and modification.

Fichiers supplémentaires

Publié-e

2022-12-03

Comment citer

1.
xavier, Gérard M, Mercier G. Recognition of Words With The Help of The Seraciroise Expert System. Canadian Acoustics [Internet]. 3 déc. 2022 [cité 7 août 2024];14(3 bis):18-9. Disponible à: https://jcaa.caa-aca.ca/index.php/jcaa/article/view/3502

Numéro

Rubrique

Actes du congrès de la Semaine canadienne d'acoustique