New AI Listens for Whale Songs in Noisy Oceans
This could help protect endangered whales from ship strikes and noise pollution.
Researchers present NAVE, a Conformer-based detector for Antarctic whale vocalization detection: temporally localizing call boundaries in long underwater recordings and classifying calls into predefined call types. NAVE combines per-bin complex mean subtraction, which suppresses constant site-specific ambient sound characteristics, with a PCEN channel that adapts to temporal variations in the noise floor.
A frequency-dynamic stem processes these noise-aware representations with convolutions that adapt across the characteristic frequency bands of target call types, and a Conformer backbone uses a convolution kernel matched to target-call duration to improve temporal localization. On BioDCASE 2026 Task 2,
- NAVE is an AI that detects whale calls in noisy underwater recordings better than previous methods.
- It filters out background noise like ship engines, making it easier to find whale sounds.
- This could help protect whales by informing ships to avoid areas where whales are present.
Why It Matters
Better whale detection can lead to fewer ship strikes and less noise pollution, helping endangered whales recover.