Generalization of Perceptual Learning of Degraded Speech Across Talkers

We investigated whether perceptual learning of noise-vocoded (NV) speech is specific to a particular talker or accent.

Four groups of listeners (n = 18 per group) were first trained by listening to 20 NV sentences that had been recorded by a talker with either the same native accent as the listeners or a different regional accent. They then heard 20 novel NV sentences from either the native- or nonnative-accented talker (test), in a 2 × 2 (Training Talker per Accent × Test Talker per Accent) design.

Word-report scores at test for participants trained and tested with the same (native- or nonnative-accented) talker did not differ from those for participants trained with 1 talker per accent and tested on another.

Learning of NV speech generalized completely between talkers. Two additional experiments confirmed this result. Thus, when listeners are trained to understand NV speech, they are not learning talker- or accent-specific features but instead are learning how to use the information available in the degraded signal. The results suggest that people with cochlear implants, who experience spectrally degraded speech, may not be too disadvantaged if they learn to understand speech through their implant by listening primarily to just 1 other talker, such as a spouse.