Contents

feature_extraction - codes (MATLAB) to extract features for the unsupervised and random forest methods
unsupervised - codes (MATLAB) for unsupervised SDM-novelty based boundary detection
random_forest - codes (Python) for the random forest implementation
cnn - codes (Python) for the CNN implementation.

For the RF and CNN methods, models trained on the 20-concert dataset are also provided, and can be used to obtain predictions on any test audio.

Description

To obtain boundary predictions for an audio file:

1. First extract all the features using the extract_features MATLAB function (more details inside feature_extraction)
2. To use the unsupervised method, run ...

3. To use the trained RF model, run (from within the random_forest directory)

predict_boundaries.py path/to/features/filename.mat

4. To use the trained CNN model, run (from within the cnn directory)

predict_boundaries.py path/to/audio/filename.wav