Speech recognition in small room reverberation: Three experimental vignettes.

Abstract

This paper describes three experiments involving normal-hearing and hearing-impaired listeners’ speech recognition in small room reverberation. A consonant–vowel–consonant–vowel nonsense syllable test (NST) was recorded in a small room with a reverberation time of 0.8 s. Experiment I compared the monaural versus binaural performance of eight normal-hearing listeners in four listening conditions of (1) control (C), (2) noise-only (N) with a +10 dB S/N (speech-spectrum noise), (3) reverberation (R), and (4) reverberation and noise (N–T). Results indicated a binaural advantage and a ranking (highest to lowest) of scores for the listening conditions: C, N, T, and NT. Experiment II involved obtaining performance intensity functions at 25, 35, 45, and 55 dB SL (re: SRT) for consonant, vowel, and bisyllable scores for ten normal-hearing listeners. While consonant and bisyllable scores were significantly better at each successive SL, vowel scores were not significantly different across the SLs. Experiment III involved completing a Sequential Information Analysis, SINFA [M. Wang and R. C. Bilger, J. Acoust. Soc. Am. 54, 1244 (1973)] on five normal-hearing (NH) and five hearing-impaired (HI) (high-frequency) listeners’ consonant responses in four listening conditions similar to Experiment I, except the N condition had +12 dB S/N with multitalker babble. Results indicated that, by far, the least amount of information was transmitted in the N–T condition (NH−1.01 bits/HI 0.61 bits) with less than 50% transmitted for most features. High-frequency features (e.g., sibilancy) fared poorly in all conditions for HI listeners. Implications of the findings will be discussed. [Work supported by a Summer Academy Research Mentorship Grant, OUHSC, Higher Regents of the State of Oklahoma.]