Fig. 4: Naïve listeners distinguish song from speech vocalizations across cultures.

A Behavioral task: 74 individuals were asked to rank, as rapidly as possible on a 5-point scale, whether each speaker was singing (code 1) or speaking code (−1), only 74 participants completed the experiment. B Behavioral ratings (chance level – 0) for song (orange) and speech (blue) samples. Diamonds represent the ratings for each listener (n = 74 independent individuals). C Fieldsites-level behavioral ratings (chance level – 0) for song (orange) and speech (blue) samples. Colored circles represent each of the 21 societies/cultures (sorted as a function of the SVM decoding accuracy of Fig. 2D - with a jet colormap - n = 21 independent societies). D Correlation between normalized difference scores (Song MPS vs. Speech MPS and Song vs. Speech behavioral ratings) represented in the MPS domain. (FDR corrected in the spectral and temporal modulation domains, p < 0.05). E Scatter plot of SVM decoding accuracy (Fig. 2D) against behavioral normalized difference (Song vs. Speech). Colored circles represent each of the 21 societies/cultures (sorted as a function of the SVM decoding accuracy of Fig. 2D, two-tailed).