fingerstory/chord-service

Finger Story — Chord Recognition Service

Wraps the BTC model (Bi-directional Transformer for Chord recognition, ISMIR 2019) as a small HTTP API. CPU-only.

Endpoints

• —GET / — health check.
• —POST /analyze — multipart form field file (audio: mp3/wav/m4a). Returns:

  { "chords": [ { "start": 2.685, "end": 4.074, "chord": "F#:7" }, ... ], "count": 42 }

Notes

• —Large-vocabulary model: 170 chord classes (maj/min, 7, maj7, min7, sus, etc.).
• —Quality ≈ Chordify level on typical pop/rock; best on clean (e.g. solo guitar) audio.
• —Deploy: Hugging Face Spaces (Docker SDK) free CPU tier, or any container host (Render/Railway/Fly).
• —The BTC code + pretrained weights are fetched and patched at build time (see Dockerfile).