Perception of fricative consonants extracted from continuous speech

Résumé fourni par la source

When listening to natural speech, listeners can integrate top-down information (such as lexical knowledge, contextual predictability, and talker characteristics) with bottom-up properties of the acoustic signal. We estimated the contribution of bottom-up properties in two identification experiments using word-internal fricative-vowel sequences (N > 700) that were extracted from the Buckeye multitalker corpus of spontaneous speech. Lab participants (N = 24) listened to stimuli presented in random order and identified each fricative as one of /f v θ ð s z ʃ h/. Fricatives, fricative-vowel combinations, and talkers were balanced within experimental lists and unpredictable. Identifications were accurate across participants (median = 0.67 [0.62–0.74]) and across fricatives (median = 0.72 [0.61–0.82]) with the exception of /ð/ (0.27). Online participants (N = 32) performed the same task with shorter lists; identifications were again accurate across participants (median = 0.67 [0.41–0.83]) and fricatives (median = 0.70 [0.56–0.88]) except for /ð/ (0.37). These results suggest that bottom-up properties provide substantial information about fricative identity in continuous speech regardless of context and talker, though they fall short of performance in a previous study that used fricative-vowel stimuli recorded in isolation (McMurray and Jongman, 2011; median = 0.93 [0.85–0.98] across participants).

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.