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

1aSC25. Quantitative measures of envelope cues in speech recognition.

Johnny Saade

Dept. of Elec. Eng., Univ. of California, Los Angeles

House Ear Inst., Los Angeles, CA 90057

Fan-Gang Zeng

John J. Wygonski

Robert V. Shannon

Sigfrid D. Soli

House Ear Inst., Los Angeles, CA 90057

Abeer Alwan

Univ. of California, Los Angeles, CA

A quantitative procedure is derived to evaluate the relative contribution of envelope cues to speech recognition. Recognition data of 16 consonants in the /aCa/ form were collected using signal-correlated noise stimuli in seven normal-hearing listeners. Several distance measures were calculated directly from duration and amplitude of the acoustic envelope. One amplitude distance measure was the Euclidean distance which was computed from the squared difference of the sample-by-sample amplitudes. The second measure was the envelope difference index (EDI) [Fortune et al., Ear Hear. 15, 93--95 (1994)] which was computed from the absolute value of the difference of the sample-by-sample amplitudes. A multidimensional scaling analysis was used to convert the perceptual confusion matrix into a distance matrix and to normalize the different distance measures. Correlation coefficients were computed between the different distance measures and the perceptual data. Preliminary analysis of data from six stop consonants showed that the consonant duration alone is sufficient to explain the perceptual data (r=0.92). Although Euclidean distance conveyed less information (r=0.75) than duration, it was a better measure than the EDI (r=0.31). Evaluation of these measures on the full 16 consonant set will be discussed.