Using the beatthis model for testing the capacity of artificial generated music as potential training data. Based on the train results it`s possible to adjust the generated music to make the train results better.
These instructions will get you a copy of the project up and running on your local machine for development and testing purposes. See deployment
Install dependencies, recommended creating a conda enviroment or pyenv.
pip install -r requirements.txt
For installing de CUDA 11.8 toolkit and related packages follow the instructions on this forum
To execute the proper preprocess from the authors, run
python launch_scripts/preprocess_audio.py
The data structure should be:
beat_this_repo/
├── data/
│ __├── audio
│ ______└── .wav audio files
│ ______└── spectrograms
│ ___________└── dataset folder
│ ___└── annotations
│ _________└── dataset folder
│ ____________└── annotations
│ _______________└── beats
│ _________________└── .beats files
│ ____________└── info.json
│ ____________└── single_or_kfold.split
│ __└── audio_paths.csv
To reproduce the train and test different datasets, use the following command
python launch_scripts/train.py
have in mind that this train does not recompile the model, for further details check the original beatthis!
However, it has not being useful recompiling the model for training purposes, even when changing the parameters.
To calculate beat inferences, use the following:
python beats_inference.py
the script will generate a beat annotation .txt or any format prefered(like .beats or other) with the beats predicted and the time they were detected.