Perception of final-consonant “voicing” in whispered speech.

Abstract

The purpose of this study was to examine the accuracy of final consonant perception in whispered speech. Of particular interest was the accuracy of the detection of “voicing” in whispered consonants and its relationship to the prominence of different acoustic cues to consonant voicing. This relationship was investigated for the effect of speaker and consonant. Stimuli were natural tokens that differed in final consonant and vowel preceding it. For example, [habæz] embedded in the carrier sentence “I said testword eight times” produced at a conversational speaking rate. Recordings from four monolingual speakers of American English were used. The consonants were two pairs of stops /b-p/, /g-k/, a fricative-pair /z-s/ and an affricate-pair d ȝ-ʈ ∫. In an eight-alternative forced choice task, listeners indicated which final consonant they heard. Analysis of the results revealed that whispered consonants were identified with a relatively high overall degree of accuracy (84%–88%). ANOVA indicated that identification accuracy was affected by both speaker and consonant. A logistic regression model was used to establish the best predictors of consonant voicing in whispered speech for classification of intended consonants and in predicting the success in [+/− voicing] detection. Predictors that were considered included several acoustic measures, e.g., vowel duration.