Divergent patterns of voicing perception in various challenging listening conditions.

Abstract

Perception of consonant voicing is especially robust, even under conditions of harsh spectral degradation, hearing loss, and background noise. Considering the limitations and variability of the acoustic input, it is unlikely that the same perceptual strategy is used in all situations. A sequence of experiments was designed to assess the weighting of multiple acoustic cues to the voicing contrast word‐initially and word‐finally, and to reveal patterns unavailable in conventional confusion matrices and information transfer analysis. In preliminary results, the combination of background noise and low‐pass filtering brought about a dramatic shift in weighting of the VOT and F0 cues for initial consonants; these degradations in isolation did not produce such a shift. Under conditions of spectral degradation (noise‐band vocoding), listeners significantly increased their use of vowel duration, and significantly decreased their use of vowel formant transition, which is dominant in the optimal condition. These results suggest that in challenging conditions, listeners are apt to use secondary acoustic cues that are potentially less reliable or less perceptually salient than primary cues used in optimal conditions. It is concluded that comparison of performances by listeners in various conditions should be guarded, even when recognition scores appear to be equivalent on the surface.