Invited talk presented at the Dept. of Linguistics, Northwestern University, May 2013.

Abstract: Many researchers have observed that speech sounds vary considerably across different contexts, an issue known as the lack of invariance. Given this variability, how much information is conveyed by individual acoustic cues? That is, how reliably do specific cues distinguish phonological contrasts? Here, I present a method for computing the reliability of acoustic cues and argue that listeners can correctly recognize speech by weighting multiple cues according to their reliability. This provides a way of overcoming the variability inherent in individual cues. I present evidence for this using several approaches: simulations with computational models, acoustic-phonetic analyses, and behavioral data from listeners. Together, the results suggest that general principles of cue integration can provide useful models for understanding how listeners recognize speech and how they weight different sources of information in the sound signal.

Talk slides: