Audio2Manus is a Python package with all the necessary tools to separate MIDI files into two tracks, one for each hand. It contains algorithms based on transcribed MIDI (output of AMT models from recordings) and also on acceleration data and video data (with the marker less tracking algorithm from the lab).
- Hand separation from Kalman Filters
- Hand separation from neural network (TensorFlow)
- Hand separation from video data
- MIDI transcription metrics
- Hand separation metrics
- MIDI concatenation
- Data Time Wrapping for MIDI synchronization
This package was tested with Python 3.9 and 3.11 with Conda.
A lot of different use cases of this package are available as jupyter notebooks on the jupyter-notebooks repository.