This project is aimed at converting timbre of a given musical piece with the use of VAEs combined with GANs.
Currently, this repository contains the automated data generation for acoustic and electric sounds that have been synthesized with the use of the fluidsynth (v1.1.6) for Windows. The data_generation/midi_generator directory contains a batch file (.bat) that, when executed, produces the following files:
generated_midi/random.midsynthesized_wav/random_acoustic.wavsynthesized_wav/random_electric.wavThis build process currently only works for windows. However, with the right fluidsynth builds and a few path adjustments, it can run on linux systems as well.
For the MIDI file generation, MIDIUtil was used. It can be installed with the following command
pip install MIDIUtil
The fluidsynth binary for x64 is already included in the data_generation/fluidsynth_binaries directory because I don't like it when I have to build everything when I download from github. I didn't use the builds for x86 because lol I didn't want to.
You will also need to download some soundfonts for this to run.
- For acoustic guitar sounds, Download the Acoustic soundfont
- For electric guitar sounds, Download the Electric soundfont
The provided batch script data_generation/midi_generation.midi_to_electric_acoustic.bat runs the data generation process and populates the synthesized_wav directory
The Jupyter Notebook has the complete implementation of the VAE.