This paper details the DS@GT ARC team's approach to BirdCLEF+ 2026, focusing on multi-label detection of animal vocalizations in soundscapes from the Pantanal wetlands. The 2026 edition adds about an hour of labeled soundscapes, shifting the task towards supervised pipelines suited for the labeled set.
First, we build a competitive supervised baseline that ensembles a frozen Perch v2 backbone, a trained HGNetV2-B0 sound-event-detection network, and a non-bird prototypical head, achieving a private leaderboard score of 0.936 at rank 1894 within a 90-minute CPU budget.
Second, we investigate whether token-based representations can compete, contrasting codec representations from neural audio codecs against semantic representations from foundational embeddings. We compare two bioacoustic specialist models against four token-based encoders trained on AudioSet. The repository for this work can be found at GitHub.
Blogger's Review: This paper highlights the competition between traditional CNNs and emerging token representations in bird sound recognition tasks. It is particularly interesting to observe how varying model architectures can achieve higher accuracy as data volume increases, emphasizing the potential and future applications of token representations.