The effect of spectral resolution on speech perception in a multitalker babble background Free

Does finer spectral resolution always have an advantage for speech recognition? In this experiment, speech was divided into separate bands within which spectral information was removed by replacing the fine structure with noise; the temporal envelope cues of each band remained intact. Normal‐hearing subjects identified consonants in a/C/a for speech and babble processed with 1, 2, 4, and 8 bands, in addition to the unprocessed speech and babble, for a range of signal‐to‐babble ratios. For high S/B ratios our results agree with earlier results, in that performance improved with increases in the number of bands, and higher transmission of information was observed for voicing and manner than place of articulation [Shannon et al., Science 270, 303–304 (1995)]. However, as S/B decreased, the advantage of better spectral resolution was reduced. This suggests that for speech in a competing babble background, increased spectral resolution not only increases the saliency of speech, but at the same time makes background babble a more effective distracter. Thus the benefit provided by fine spectral resolution depends upon the task and S/B ratio. [Work supported by NIDCD.]