Investigation of the neural bases of context-dependent speech categorization

Previous research has demonstrated that simple sequences of preceding sine-wave tones affect speech categorization in a spectrally-contrastive manner [L. Holt, Psychol. Sci. 16, 305–312 (2005)]. The current research explicitly links these effects to effects commonly thought to be instances of talker normalization [P. Ladefoged and D. E. Broadbent, J. Acoust. Soc. Am. 29, 98–103 (1957)]. Synthesized sentences manipulated to sound like different talkers influence categorization of a subsequent speech target only when sentences’ long-term average spectra (LTAS) predict spectral contrast. Likewise, sequences of tones modeling these LTAS differences produce parallel context-dependent speech categorization effects. The predictiveness of LTAS, rather than perceived talker, suggests that general auditory rather than speech-specific or articulatorily-driven mechanisms may play a role in effects considered to be instances of talker normalization. The behavioral measures are paired with magnetoencephalography (MEG) to investigate the neural bases of these parallel effects. Listeners categorized a /ga/-/da/ series in the context of preceding sentences and nonspeech tone sequences varying in their LTAS while MEG signals were acquired. Analyses focus on how the speech target is encoded as a function of preceding LTAS and the status of the context as speech or nonspeech. [Work supported by NIH and NOHR.]