MuseBridge
Song recommendations based on how the lyrics feel
Recommends songs by emotional similarity instead of genre. Every song in a large Spotify tracks dataset gets an emotion vector from its lyrics, and nearest neighbours in that space become the recommendations.
Highlights
- Async lyrics fetcher (aiohttp, LRCLIB API) with bounded concurrency over the full dataset
- DistilRoBERTa emotion model run in GPU batches to build a 7-emotion vector per song
- KNN recommender with a Flask web UI and a CLI