Distributed hierarchical filters partition the Danionella vocal repertoire

Vocal repertoires consist of a limited number of call types, even though their sounds can vary continuously in form. Where in the brain receivers partition this variation into discrete types is contested: different accounts place the emergence of categories in the cortex/pallium or already in the midbrain. Because recordings have been limited to isolated regions, it remains unknown whether this partitioning is confined to a single region at all, occurs in parallel across many, or is built up along the pathway. Here we present the first whole-brain, cellular-resolution analysis of how conspecific vocalizations are categorized in a vertebrate, using the glassfish Danionella cerebrum. We find that categorization is neither pallial nor localized. It begins at the first central auditory nuclei, much earlier than previously thought, where a continuous morph between social and non-social sounds evokes an abrupt switch in population activity. The midbrain then sharpens these representations, and the thalamus filters for species-typical pulse rates while splitting call duration into two discrete populations. Only downstream of this shared code do the sexes diverge in forebrain social nuclei. Call type partitioning is thus distributed and hierarchically assembled, rather than read out at any one stage.