ShaRinGan

DCGAN that converts raw guitar sound into effector modified sound.

Getting Started

Clone this repository

git clone https://github.com/hamling-ling/ShaRinGan.git

Download training data set. This takes about 30 sec.

$cd ShaRinGan/data/raw_waves
$./download.sh

Convert wave files to bin files

$cd ../../src
$python create_waves_wavfiles.py

Run Training(you can skip)

$cd ../../src
# this takes a few hours with GPU
$python sharingan_train.sh

Or Download pre-trained model

$cd ../../data/output
$download_pretrained.sh

Run Inference to convert raw to effectored suond

$cd src
$./sharingan_cvt.sh

You will see raw input, ground truth and converted wav files

$ls ../data/cvt/pretrained_model/
input.wav  output.wav  target.wav

• input.wav as an input source.
• target.wav is a ground truth.
• output.wav is what generated by GCGAN.

Visualize

Run visualizing script

$cd src
./sharingan_test.sh

Then you'll get bunch of images like this. Note that blue: input source, green: ground truth and orange: inference.

Prerequisites

• tensorflow 1.13.1
• pysoundfile

Jetson Nano Application

Guitar Effector application using this model can be found in app directory. "Jetson Nano DCGAN Guitar Effector"

License

This project is licensed under the MIT License - see the LICENSE.md file for details