SpecGAN_DatasetGeneration
Dataset generation for training a SpecGAN
This script takes a folder full of audio files and extracts harmonic and percussive samples, creating a dataset useful for training a SpecGAN based on your favorite music.
I've used Librosa Harmonic/Percussive separation, then I adapted this script for samples separation based on onset detection.
I then used this repo for training my SpecGAN.
Usage
python genDataset.py -i myAudioDir/ -o myOutputDir/
Samples
Here the samples dir with some percussive examples extracted from this song.