Internal representation of spectral information in dependence of signal type and auditory filter bandwidth

Abstract

Previous studies have shown that enhancement of the spectral background (lower level spectral components) improves intelligibility as well as listening comfort of music signals in subjects with broader than normal auditory filters. This is in contrast to reported results on speech perception in noise, yielding reduced intelligibility as a result of spectral contrast reduction. A proposed model for the internal object representation by separation of ‘‘spectral layers’’ accounts for these opposing effects. ‘‘Spectral layers,’’ corresponding to different sound sources [in music: dominant/accompanying musical voices; in speech-in-quiet: formant peaks/modulation sidebands of the peaks (transitions); in speech-in-noise: speech foreground/speech background/noise components] are represented in coherent spectral amplitudes. The audibility of ‘‘spectral layers’’ is derived from individual masking properties. Pilot experiments with speech tokens and short music fragments agree relatively well with the model predictions: The effect of either increasing or reducing the spectral contrast on discrimination highly depends on the distribution of the perceptually relevant acoustical information in the ‘‘spectral layers’’ relative to their audibility. Performance varies for different signal types (simultaneous musical voices, vowel, consonant). [Work supported by the Austrian Academy of Sciences.]