Extraction of vowel information from fricative spectra

Abstract

Although steady-state spectra can specify fricative identity, there are variations in the spectra of fricatives produced in different phonetic contexts. We examined whether such variations in initial fricatives of fricative-vowel-fricative-vowel-fricative utterances contain sufficient information for identification of the subsequent vowel. Twelve utterances made by combining fricatives /s. ∫. z. ȝ/ with vowels /i, a, u/, e.g., /sasas/, were recorded from two speakers. Using a computer controlled editing procedure, initial fricative segments of 150 ms, measured back from vowel onset, were excised from each utterance. These segments, equated for amplitude, comprised the stimuli for identification tests. Eight listeners were asked to identify the vowel which had followed each fricative segment in the original utterance, while a second group was asked to identify the fricative. Both groups received an initial identification test, identification tests with feedback, and a final identification test. Initial results show that vowel identification is above chance and improves with feedback, although not to the level of fricative identification. This suggests that sufficient information is present in initial fricative spectra to specify the subsequent vowel.