Using statistical decision theory to predict speech intelligibility. III. Effect of audibility on speech recognition sensitivity

摘要

The speech recognition sensitivity (SRS) model [H. Musch and S. Buns, J. Acoust. Soc. Am. 109, 2896-2909 (2001)] is a macroscopic model for predicting speech intelligibility. The present study proposes a modification to the relation between the SRS model's audibility-noise variance and the signal-excitation to noise-excitation ratio (SNRE) in the auditory periphery. The modified relation is derived from data obtained in nine studies that measured normal-hearing listeners' consonant-recognition performance at several levels of speech-spectrum shaped noise. When the audibility-noise variance is directly proportional to the relative power of the noise excitation in the auditory periphery, the SRS model yields good predictions of the data. Four of the nine studies also reported consonant-recognition performance in various filtering conditions. Good predictions of these data were achieved with SRS model parameters that were consistent with the model parameters fitting the speech-in-noise data and with the model parameters used in the original SRS papers. (C) 2004 Acoustical Society of America.