Figure 1

Schematic view of the methodology used to investigate the mental representations associated with specific sound concepts for different populations. (A) We collected ratings on an orchestral sound dataset using four sound concepts from three participant groups with the Best–Worst Scaling methods. (B) Using these ratings, we computed consistency metrics and measured similarities and differences between groups and between concepts. (C) We extracted acoustic features (e.g., spectral centroid, attack slope) and meta features (i.e., instrument, playing technique) from the sound dataset. (D) We trained a tree-based model and assessed the most important features for the prediction of the model.