George Vengrovski

The paper trains a self-supervised bioacoustic encoder using labels from a vision-language model on citizen-scientist recordings, achieving better performance than human-annotated supervised models and YOLO detectors.

The paper shows that lesions in the medial part of the Anterior Forebrain Pathway in adult canaries cause stutter-like behavior in song, indicating its role in controlling phrase duration.

SongMAE improves birdsong analysis by using a high-resolution masked autoencoder with custom temporal and frequency masking, outperforming existing encoders in syllable classification and parsing.

TweetyBERT is a self-supervised transformer model that parses birdsong without human labels, learning acoustic and temporal patterns in canary song.