ASA 127th Meeting M.I.T. 1994 June 6-10

4pSP26. Effects of speaker, token, and noise variability on the perception of vowels in noise.

Res. Lab. of Electron., 36-747, MIT, Cambridge, MA 02139

In tests of speech perception, performance decreases when the stimulus set consists of multiple speakers compared to single-speaker sets [e.g., Mullenix et al., J. Acoust. Soc. Am. 85, 365--378 (1989)]. Similarly, given a single speaker, performance decreases when the stimulus set consists of multiple tokens compared to single-token sets [e.g., Reed et al., J. Rehab. Res. Dev. 28, 67--82 (1991)]. This experiment examined the effect on speech perception when stimulus sets consisted of both multiple speakers and multiple tokens. Listeners identified nine vowels in /bVt/ context masked by noise at a constant instantaneous signal-to-noise ratio. One set of stimuli consisted of three speakers, each producing three tokens. Another set consisted of three speakers, each producing one token, with three different noise samples masking each token. In the baseline conditions, the sets were randomly ordered. In the experimental conditions, the sets were blocked by token, by speaker or noise, or by both token and speaker or token and noise. Averaged across listeners, the effect of token and speaker variability were additive, as were the effects of token and noise variability. However, individual listeners did not always show additivity of variability effects. [Work supported by NIH.]