MetaPerch: Learning from metadata for bioacoustics foundation models
The model trains on recording metadata as an extra signal, not just animal sounds.
MetaPerch uses details such as location and time from citizen-science bioacoustic datasets as auxiliary supervision. The authors say those signals help the model learn species-metadata correlations that can improve robustness under species distribution and acoustic domain shifts. They report strong species identification across multiple challenging domains, with experiments covering 9 metadata sources and 17 datasets. ArXiv · AI/CL/LG's note
MetaPerch uses details such as location and time from citizen-science bioacoustic datasets as auxiliary supervision. The authors say those signals help the model learn species-metadata correlations that can improve robustness under species distribution and acoustic domain shifts. They report strong species identification across multiple challenging domains, with experiments covering 9 metadata sources and 17 datasets. ArXiv · AI/CL/LG's note
score 4