Muralidhar Kudumala
Elec. Eng., Univ. of Oklahoma, Norman, OK
Blas Espinoza-Varas
Univ. Oklahoma Health Sciences Ctr., Oklahoma City, OK 73190
This study examines some advantages of adaptive psychophysical tests for obtaining reliable and efficient estimates of speech quality. Specifically, in vowels corrupted by additive noise, noise-detection thresholds were measured adaptively. The 350-ms vowels were obtained by addition of the first 32--35 harmonics of 100--200 fundamental frequencies, with spectral envelopes appropriate for /i/ and /a/. Noise was synthesized by random-phase addition of harmonics of a 10-Hz fundamental; the noise bandwidth and the spectral envelope were equal to those of the vowels. In a 2IFC task, subjects had to determine the observation interval that contained a ``noisy vowel.'' During a block of trials, the noise level was varied using a ``two-down, one-up'' rule (Levitt, 1971); the test terminated when the subject completed 6--8 reversals, and thresholds were defined as the average noise level obtained in the last three reversals. The adaptive method yielded threshold estimates in much fewer trials than a constant stimuli procedure used in previous studies. The adaptive method was implemented in labview software, National Instruments DSP boards, and a Macintosh computer. [Work supported by OCAST.]