ASA 129th Meeting - Washington, DC - 1995 May 30 .. Jun 06

5pSC31. Perceptual normalization of synthesized vowels following natural speech and breath sounds.

Sonya M. Sheffert

D. H. Whalen

Haskins Labs., 270 Crown St., New Haven, CT 06511

The identification of synthetic vowels relies in part on a precursor phrase, suggesting that normalization is taking place. Breath intake sounds, which can also precede speech, may also give information. This study examines both speech and breath precursors. Two adult speakers (one male, one female) produced the phrase ``The next one is.'' They also produced audible breath sounds before saying the words ``bead,'' ``bad,'' or ``bud.'' Two 5-member continua were created, one from ``bid'' to ``bed,'' another from ``bad'' to ``bud,'' based on these words produced by each speaker. Formants were averages (across the two speakers) for the natural productions; F0 was 160, resulting in gender neutral tokens. The gender of precursor speech and breath before each continuum item were varied. The vowel information in the breaths also favored one end of the continuum or the other. Two results suggest that, under some circumstances, vowel identification was affected by breath precursors: (1) There were more /(small capital eye)/ responses on the ``bid-bed'' continuum following the female breaths; (2) there were more /(ae ligature)/ responses following the male and female (ae ligature)-breath. Paradoxically, natural speech precursors shifted identification toward the opposite gender's formant space. Another test suggests that these vowels identification shifts did not arise postperceptually.