Researchers Develop TweetyBERT To Parse Birdsong

University of Oregon researchers publish TweetyBERT in the journal Patterns, a self-supervised model that automatically segments canary songs into notes, syllables, and phrases. The transformer-based tool matches expert annotators on 30–40-syllable canary sequences without human labels, dramatically speeding annotation. It enables large-scale, longitudinal studies in neuroscience and scalable bioacoustic monitoring for ecological and conservation research.