Upload an audio file (max 30 seconds).
GNN infers on new recordings directly, by building a representation from the recording's own acoustic features via message passing.