Channel segregation improves perception of speech with temporally desynchronized bands.

Abstract

Stilp et al. [J. Acoust. Soc. Am. 122, 2971 (2007)] investigated the intelligibility of temporally desynchronized bands of speech and concluded that listeners’s resilience to temporal asynchrony may possibly be due to differences in fundamental frequency (f0) across syllables which may help in segregating bands. This hypothesis is tested in two experiments in temporally desynchronized bands but with spectral manipulations designed to either aid or inhibit stream segregation. Seven‐syllable sentences were synthesized at three different speaking rates and processed by four nonoverlapping 1/3‐octave filters. Onsets of the lowest‐ and highest‐frequency bands were parametrically delayed. In Experiment 1, f0 in delayed bands was uniformly elevated using pitch‐synchronous overlap add synthesis. In the control condition (no f0 manipulation), intelligibility was nonmonotonic with delay across speaking rates with local minima corresponding to the duration of one syllable. The two‐speaker manipulation decreased spectral similarity across bands making intelligibility more resilient to temporal distortion. In Experiment 2, f0 contours were flattened. Intelligibility was uniformly poorer than in the control condition; the increased spectral similarity compromised listeners’ ability to segregate information from band pairs at various delays. Acoustic measures of potential information, absent explicit linguistic information, reinforce the strong relationship between spectral predictability and intelligibility. Overall, results provide support that channel segregation plays an important role in the perception of temporally desynchronized bands. [Supported by NIDCD]