Wav2Vec2 pretrained on 5 years of dolphin recordings beats general bioacoustic models
emaros96 · x · 2026-10-22
After five years of recordings, a Wav2Vec2 model pretrained on a large dolphin vocalization corpus beats general bioacoustic models (AVES, BioLingual) on both evaluation tasks, with plenty of room left to improve. The dataset, code and model are released on GitHub and Hugging Face.
More from Research
- ToM2Vec encodes Theory of Mind as vectors, beating baselines in Overcooked agent QA — EliasEskin · 2026-10-22
- CONCERT: Generative Modeling of Spatially Resolved Perturbation Responses — anshulkundaje · 2026-10-22
- ToM2Vec trained via self-play with Qwen3-8B transfers to GPT-5.4 teams, adding 17.3 points — EliasEskin · 2026-10-22
- Probing ToM2Vec: latent vectors beat baselines on belief, desire and action prediction — EliasEskin · 2026-10-22
- ToM2Vec uses continuous theory-of-mind vectors to boost multi-agent coordination across 5 tasks — EliasEskin · 2026-10-22
- Latent beats verbal: ToM2Vec outscores language ToM baseline by 9 points on Overcooked — EliasEskin · 2026-10-22