Navigation auf


Department of Computational Linguistics

Perceptual generalization in adapting to novel speakers


Listeners rapidly adapt to new speakers despite considerable variability in the speech patterns from one speaker to the next. One broad perceptual mechanism that could facilitate this process is one of perceptual generalisation: that is, listeners may generalise speaker-specific patterns from the speech sounds they have heard to speech sounds that have not yet occurred in the exchange. For instance, a speaker with a long VOT for /p/ is likely to also have a long VOT for /k/; in turn, a listener tends to associate a long, rather than short VOT for /k/ with the speaker after hearing that the speaker also has a long VOT for /p/. This project investigates perceptual generalisation and how it may differ from general auditory mechanisms of adaptation such as “spectral contrast”. Examples of phonetic dimensions that can be investigated include sibilant spectral properties, stop voice onset time, and potentially vowel formants.