Keynotes | K&Ts | GACs | Talks | Posters | Search

Poster F in Poster Session F: Thursday, August 6, 1:45 – 3:30 pm, Kimmel Center, Shorin & Rosenthal Rooms

Linear spectrotemporal tuning accounts for much of phonetic feature encoding in the human auditory cortex

Krish G. Patel¹, Guoyang Liao¹, Dana Boebinger¹, Kirill V Nourski², Matthew Howard², Christopher Garcia², Thomas Wychowski¹, Webster Pilcher¹, Samuel Victor Norman-Haignere¹; ¹University of Rochester, ²University of Iowa

Presenter: Krish G. Patel

Speech perception is challenging because features such as phonemes are only partially correlated with acoustic structure, making them difficult to infer from the speech signal. Our understanding of speech encoding in human auditory cortex has been shaped by studies asking whether speech features, like phonemes or phonetic features, predict neural responses beyond an acoustic baseline. Baseline choice is thus critical: it determines which aspects of neural responses are attributed to simple acoustic tuning and which are interpreted as going beyond it. Here we examine how baseline choice influences estimates of speech-feature encoding using spatiotemporally precise intracranial recordings from human auditory cortex during acoustically diverse spoken sentences. We find that phonetic feature models explain substantial unique variance relative to a one-dimensional waveform envelope baseline, particularly in speech-selective regions of non-primary auditory cortex. However, phonetic features explain little variance beyond linear spectrotemporal tuning, whereas spectrotemporal tuning explains substantial variance beyond phonetic features, even in speech-selective regions. These results indicate that much of the variance predicted by phonetic features can be captured by simple linear spectrotemporal computations, highlighting the acoustic baseline as central to interpreting speech encoding models.

Topic Area: Auditory, Speech & Language Processing