Effects of context and frequency shifts in vowel identification

Abstract

A high-quality vocoder (Kawahara’s STRAIGHT) was used to study the effects of frequency shifts on the identification of vowels in syllables along a continuum from ‘‘bit’’ to ‘‘bet.’’ Natural speech versions of ‘‘bit’’ and ‘‘bet’’ were recorded by an adult female talker in the carrier sentence, ‘‘Please say the word —again.’’ Synthesized versions were constructed with upward and downward shifts in formant frequencies (FF; scale factors 0.9, 1.0, or 1.1) in combination with shifts in fundamental frequency (F0; scale factors 0.8, 1.0, or 1.2). When the carrier sentence and target syllable were both shifted, listeners reported more ‘‘bit’’ responses as the F0 shift increased and the FF shift decreased. Systematic shifts in the identification functions were also found when only the carrier was shifted. Overall, the results are consistent with a model that simulates the categorization responses of human listeners based on statistical distributions of acoustic measurements in natural speech.