Talker normalization and acoustic properties both Influence spectral contrast effects in speech perception

Abstract

Talker normalization (TN) occurs when listeners adjust to a talker’s voice, resulting in faster and/or more accurate speech recognition. Several have suggested that TN contributes to spectral contrast, the perceptual magnification of changing acoustic properties. Studies using sine tones in place of speech demonstrated that talker information is not necessary to produce spectral contrast effects (Laing et al., 2012 Front. Psychol.). However, sine tones lack acoustic complexity and ecological validity, failing to address whether TN influences spectral contrast in speech. Here we investigated how talker and acoustic variability influence contrast effects. Listeners heard sentences from a single talker (1 sentence), HINT (1 talker, 200 sentences) or TIMIT databases (200 talkers, 200 sentences) followed by the target vowel (varying from “ih” to “eh” in F1). Low (100–400 Hz) or high (550–850 Hz) frequency regions were amplified ( + 20, + 5 dB) to encourage “eh” or “ih” responses, respectively. When sentences contained + 20 dB spectral peaks, contrast effect magnitudes were comparable across conditions. When sentences contained + 5 dB peaks, contrast effect magnitudes decreased overall, but were smallest following TIMIT sentences with larger, comparable effects for single-talker conditions. Thus, TN influences contrast effects when spectral peaks are modest, but not when they are large.