Characterizing Vocal Repertoires—Hard vs. Soft Classification Approaches
(A) Number of clusters in dependence on the fuzziness parameter μ. Partitions with more than five clusters exist only over very narrow ranges of μ values (red). (C-D) membership matrices for the identified clusters: Rows correspond to different fuzzy clusters and columns to individual calls. Membership values of single calls to each class are color coded (B). The scream-cluster is the first to emerge (cluster 1, C), followed by grunts (cluster 2, D). The scream-cluster splits into two clusters and the weaning-cluster emerges (cluster 1–2; cluster 4, E).