Intelligibility of locally scrambled mosaic speech

It has been well-known that locally time-reversed speech of meaningful sentences is perfectly intelligible when the segment duration is shorter than or equal to 40 ms. However, how the auditory system can use temporally displaced speech cues with degraded amplitude envelopes has been largely unknown. Here, we show that the intelligibility of mosaic speech, i.e., spectrotemporally degraded noise-vocoded speech produced with critical bands and stepwise amplitude envelopes, dropped from 100% to 1% by extending the segment duration from 20 to 320 ms. However, mosaic speech with a 20-ms segment duration retained relatively high intelligibility (100%, 98%, 76%, 62%, and 13%) when the temporal order of mosaic segments was scrambled within 40-, 80-, 120-, 160-, and 320-ms spans. Extending mosaic segments to 40 ms almost did not affect intelligibility benefits. Scrambling the temporal order independently across frequency bands almost completely removed these intelligibility benefits. These results suggest that the auditory system can use relevant speech cues to some extent despite the temporal order randomization and pitch removal, provided that the speech cues retain coherence across frequency. The results also support the two-temporal-windows integration (2TWI) model for speech perception, proposed by D. Poeppel and his colleagues.