Can Tokens Compete? Token Representations against Supervised CNN Backbones for BirdCLEF+ 2026

2026-07-15

AI Summary

This study addresses the challenge of multi-label animal vocalization detection in the soundscapes of the Pantanal wetland under limited labeled data. The authors construct a supervised baseline that integrates a frozen Perch v2 backbone, an HGNetV2-B0 sound event detection network, and a non-avian prototype head. For the first time in the BirdCLEF+ 2026 task, they systematically compare the performance of token representations derived from neural audio codecs against semantic embeddings from foundation models within a supervised learning framework, proposing a novel and efficient paradigm that combines supervised models with token-based encoders. Experimental results demonstrate that the proposed approach achieves a score of 0.936 on the private leaderboard within a 90-minute CPU constraint, validating the effectiveness and potential of specialized token representations for bioacoustic tasks.

Abstract

This paper details the DS@GT ARC team's approach to BirdCLEF+ 2026, 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 toward supervised pipelines fit to 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, reaching a private leaderboard score of 0.936 at rank 1894 within a 90-minute CPU budget. Second, we ask 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 https://github.com/dsgt-arc/birdclef-2026.

Problem

• multi-label detection
• animal vocalizations
• token representations
• supervised CNN backbones
• soundscapes

Innovation

• token representations
• supervised CNN backbones
• neural audio codecs
• foundational embeddings
• bioacoustic detection