Fig. 5: Superiority of STM features over other acoustical variables.
From: Spectro-temporal acoustical markers differentiate speech from song across cultures
A R values in the STM domain for the correlation between fitted responses (PLS) and observed responses (STM) (FDR corrected in the spectral and temporal modulation domains, p < 0.05, two-tailed). Dark lines illustrate the boundaries of the significant effects presented in (Fig. 2A). B, C Left Panels: Scatter plot of fitted responses (PLS) and observed responses (STM) for the spectral (see letter (S) in (A), statistical peak (B), and temporal (see letter (T) in (A) statistical peak (C)). Circles represents each speakers/vocalization (n = 369), two-tailed, all ps < 0.001. Right panels: VIP scores: the horizontal bars show the acoustic features with the largest influence in the PLS. D Fieldsite-wise cross-validated (n = 21 independent societies) support vector machine decoding accuracy (chance level: 50%) for four alternative strategies with STM features only (white), STM + acoustical features (black), Acoustical features only (gray) and Acoustical features without VIP variables (light gray). The colored dots (jet colormap sorted as a function of accuracy for the model trained with STM features only) represent the accuracy for each society. **p < 0.001, ns non-significant, post hoc pairwise comparisons were two-tailed, Bonferroni corrected.
