Music transcription is the task of transcribing piano recordings into MIDI files. This repo was forked from https://github.com/qiuqiangkong/piano_transcription_inference and has the code and trained models for our research project for UBC CPSC 532S deep learning course!
All details available in our report, available in this project_report folder.
Demos: Coming soon
Some MIDI output examples available in the sample_results folder.
Install dependencies:
pip install -r requirements.txt
We trained on the MusicNet dataset for the multi-instrumental model and on the MAESTRO datset for the piano model.
Some additional code for self-attention blocks was adapted from https://github.com/sahajgarg/image_transformer