Omnizart is a Python library that aims for democratizing automatic music transcription. Given polyphonic music, it is able to transcribe pitched instruments, vocal melody, chords, drum events, and beat. This is powered by the research outcomes from Music and Culture Technology (MCT) Lab.
Visit the complete document for detailed guidance.
# Install omnizart
pip install omnizart
# Download the checkpoints
omnizart download-checkpoints
# Transcribe your songs
omnizart drum transcribe <path/to/audio.wav>
omnizart chord transcribe <path/to/audio.wav>
omnizart music transcribe <path/to/audio.wav>docker pull mctlab/omnizart:latest
docker run -it mctlab/omnizart:latest bash
| Application | Transcription | Training | Evaluation | Description |
|---|---|---|---|---|
| music | ✔️ | ✔️ | Transcribe musical notes of pitched instruments. | |
| drum | ✔️ | Transcribe events of percussive instruments. | ||
| vocal | ✔️ | ✔️ | Transcribe note-level vocal melody. | |
| vocal-contour | ✔️ | ✔️ | Transcribe frame-level vocal melody (F0). | |
| chord | ✔️ | ✔️ | Transcribe chord progressions. | |
| beat | ✔️ | ✔️ | Transcribe beat position. |
NOTES The current implementation for the drum model has unknown bugs, preventing loss convergence when training from scratch. Fortunately, you can still enjoy drum transcription with the provided checkpoints.
Currently, Omnizart is incompatible for ARM-based MacOS system due to the underlying dependencies. More details can be found in the issue #38.