Process individual tracks

Hit the button to process tracks online!

To process a piano trio recording using our pipeline, you can use a command line interface to run the code in src/process.py. For example, to process 30 seconds of audio from Chick Corea’s Akoustic Band ‘Spain’:

git clone https://github.com/HuwCheston/Jazz-Trio-Database.git
cd Jazz-Trio-Database
python -m venv venv
call venv/Scripts/activate.bat    # Windows
source venv/bin/activate    # Linux/OSX
pip install -r requirements.txt
bash postBuild    # Linux/OSX only: download pretrained models to prevent having to do this on first run
python src/process.py -i "https://www.youtube.com/watch?v=BguWLXMARHk" --begin "03:00" --end "03:30"

This will create a new folder in the root directory of the repository: source audio is stored in /data, annotations in /annotations, and extracted features in /outputs. Extracted features follow the format given in Cheston, Schlichting, Cross, & Harrison (2024).

By default, the script will use the parameter settings described in Cheston, Schlichting, Cross, & Harrison (2024) for extracting onsets and beats. This can be changed by passing -p/--params, followed by the name of a folder (inside references/parameter_optimisation) containing a converged_parameters.json file.

The script will also use a set of default parameters for the given track (e.g. time signature). To override these, pass in the -j/--json argument, followed by a path to a .json file following the format outlined in the metadata table above.