Extrinsic talker normalization via rapid accumulation of talker-specific phonetic detail

Abstract

Talker normalization makes speech processing more efficient by reducing the degrees of freedom between the highly variable acoustic realization of speech and its abstract mental representation. However, it is unknown how processing gains afforded by talker normalization depend on the amount of talker-specific information available in the preceding speech context. We explored the timecourse of extrinsic talker normalization in a series of speeded classification tasks that parametrically varied the amount of talker-specific phonetic detail available in the preceding speech context. In single- and mixed-talker conditions, listeners identified a target word (either “boot” or “boat”) preceded by either no speech context, a short carrier phrase (“It's a...”), or a long carrier phrase (“I owe you a...”) spoken by the same talker as the target word. Listeners' word identification was always slower in the mixed-talker condition than the single-talker one. However, the amount of interference decreased as the amount of preceding speech context increased: the interference effect of the mixed-talker condition was greatest with no carrier, less with a short carrier, and least in the long-carrier condition. These results reveal how extrinsic talker normalization facilitates speech processing via rapid accumulation of talker-specific detail from the speech context.