Midi Transcription

An implementation of Score Transformer: Generating Musical Score from Note-level Representation

An attempt at transcribing midi notes to a musicxml file

Instructions

Clone this repository:

git clone https://github.com/BarbosaRT/Midi-Translation.git

Access the directory:

cd Midi-Translation

Now install the dependencies:

pip install -r requirements.txt

To process the datasets included:

python3 main.py

If you want to train:

onmt_build_vocab -config mid_mxl.yaml -n_sample 40000 

onmt_train -config mid_mxl.yaml 

If you just want an Inference with the existing model:

• Download the model's weights here and unzip the run folder inside the datasets directory
• Change the song file in tokenize_a_song.py, then:

python3 tokenize_a_song.py  

onmt_translate -model datasets/run/model_step_80000.pt -src tokens.txt -output pred.txt -gpu 0 -verbose  

python3 song_combiner.py   

Notes

If you want to improve this code, feel free to do it. The trained model is ineffective at transcribing probably due to the lack of a large corpora of musical scores and also some problems with converting to tokens, currently it doesn't support bpm changes it defaults the song's bpm to 120.

Citation:

Score Transformer: Generating Musical Score from Note-level Representation

@inproceedings{suzuki2021,  
 author = {Suzuki, Masahiro},  
 title = {Score Transformer: Generating Musical Score from Note-level Representation},  
 booktitle = {Proceedings of the 3rd ACM International Conference on Multimedia in Asia},  
 year = {2021},  
 pages = {31:1--31:7},  
 doi = {10.1145/3469877.3490612}  
}