Music Genre Classification with Tensorflow

Replicating parts of Schindler's paper on Comparing Shallow versus Deep Neural Network Architectures for Automatic Music Genre Classification.

For COMSM0018, designed to be run efficiently on the University of Bristol's BC4 supercomputer.

Usage

python main.py [options]

command line options are:

• --depth [shallow or deep] (default shallow)
• --epochs [int] (default 100)
• --samples [int] (default 11250)
• --augment [bool] (default False)
• --batch_normalisation [bool] (default False)
• --batch_size [int] (default 16)

For example, to run a deep network with data-augmentation, with only 20 epochs and 1000 audio samples: python main.py --depth deep --augment True --epochs 20 --samples 1000

Requirements

• Tensorflow 1.6
  • (The following is required for running on Blue Crystal)
    module add languages/anaconda2/5.0.1.tensorflow-1.6.0
• librosa