A terminal tool for calculating SDR (Signal-to-Distortion Ratio), designed to quickly evaluate the performance of music separation models.
This tool provides a convenient way to assess the quality of music separation models by calculating the SDR between reference and estimated audio files.
git clone https://github.com/AliceNavigator/sdr-eval.git
cd sdr-eval
conda create -n sdr-eval python=3.10
conda activate sdr-eval
pip install -r requirements.txt
python sdr.py [-h] [--reference_folder REFERENCE_FOLDER] [--estimated_folder ESTIMATED_FOLDER] [--num_workers NUM_WORKERS]
-h, --help
: Show this help message and exit--reference_folder REFERENCE_FOLDER
: Folder containing reference audio files to process--estimated_folder ESTIMATED_FOLDER
: Folder containing estimated audio files to process--num_workers NUM_WORKERS
: Number of workers used for SDR computation
python sdr.py --reference_folder ./reference_audio --estimated_folder ./estimated_audio --num_workers 4