Fig. 2: Mandarin lexical tones evoke differential neural responses at single STG electrodes, explained by tuning to speaker-normalized pitch features.

a Electrode locations for one participant. Dark red indicates electrodes that were responsive to speech. Circled orange electrodes had significantly different response patterns to different tones. b The actual high-gamma (HG) responses from a single example electrode (Electrode 1 in a) that differentiates lexical tones. Average neural activity (shaded area represents mean ± s.e.m.) in response to each tone is plotted, aligned to vowel onset. Black lines indicate time points where means are significantly different between tones (P < 0.05, F test, two-tailed, Bonferroni corrected). c The predicted responses from an encoding model using spectrum, intensity and absolute pitch features. d The predicted responses by adding tone category features to the model in c. e The predicted responses from the full encoding model where speaker-normalized pitch features are included. f The temporal receptive field (regression weights) from the model with regard to absolute pitch, relative pitch height and pitch change, indicating selectivity to positive pitch change. g–k Same as b–e, but for Electrode 2, which is tuned to negative pitch change.