Processus d'extraction de traits pertinents dans le traitement de données acoustiques

Abstract
This research deals with analogy conversion processes of auditory data into distinctive infra-phonemic unities. These processes have been observed both in tasks of noises verbal categorization and in tasks of word or phoneme recognition in reading with an auditory priming. These conversion processes of acoustic data into phonetic data are in a proeminent position through their influence in the orientation of treatments implicated in these reading tasks. Tasks of lexical decision and phoneme monitoring on a computer in acoustic priming situation (subjects wear a headphone and simultaneously hear a 50 ms natural noise when a word or pseudo-word appears on the screen), show that analogy conversion processes of pertinent features in the prime noise can guide bottom-up treatments appearing in these experiments (lexical units influence on low levels of target stimuli treatment). Moreover, this extraction is even efficient in an irregular distribution of the acoustic energy (natural noises).