Music boundary detection using U-Net

This works for SALAMI v2.0 dataset.

Hyperparameters and configurations are in hparams.py

Feature Extraction

python feature_extraction.py KIND_DATA [--mode=MODE] [--num-workers=N]

• KIND_DATA can be 'train' or 'test'.
• MODE can be 'io', 'in', or 'out' (means what feature will be processed). Default is 'io'.
• N can be an integer from 1 to cpu_count(). Default is cpu_count() - 1.

Creation of Test set

By match_salami_file_struct.py, mp3 audio files and annotation text files are saved as SALAMI files.

DNN Training and Testing

python train_test.py [--test=EPOCH] [--(hyperparameter name)=(python script or str)]

Test Result Analysis

python analyze_test.py EPOCH [--song={ID1, ID2, ...}]

• ID1,ID2,... are song ids to be plotted in forms of mel and boundary.

Requirements

• python >= 3.7 (or 3.6 with dataclasses backport)
• numpy
• matplotlib
• PyTorch >= 1.0
• tensorboardX >= 1.7
• librosa
• tqdm
• torchsummary
• mir_eval