Level-dependent effects on speech intelligibility

Abstract

The effect of noise on speech intelligibility is typically measured using fixed-level speech (or noise) and varying the speech-to-noise ratio (SNR). An assumption of this procedure is that intelligibility mostly depends on the SNR and barely depends on speech level. The effective SNR, however, (i.e., the SNR in the internal stimulus representation), possibly depends on peripheral compression. Indeed, compression could facilitate and hinder intelligibility for negative and positive SNRs, respectively. Insofar as compression varies with level, speech intelligibility might also vary with speech level. Here, we tested these hypotheses by measuring percent correct digit triplet identification as a function of speech level for fixed SNRs. Measurements were carried out for normal-hearing subjects and for hearing-impaired subjects with linear cochlear responses, as assessed using the temporal masking curve method. Results for both groups suggest that the detrimental effect of the noise on intelligibility is larger for speech levels near threshold, particularly for negative SNRs, a result that cannot easily be explained by compression. Alternative explanations for the result are discussed.