jerrygood0703 / speech-enhancement-WGAN

speech enhancement GAN on waveform/log-power-spectrum data using Improved WGAN

SEGAN with improved wgan

Under construction!! Don't fork!!

Tensorflow 1.2rc

Using imporved wgan

Enhancement on both waveform data and LPS data

Stage1 training only L1/L2 loss, without adversarial loss

Stage2 joint training

Usage

Preparing data(data_utils.py)

import tensorflow as tf
from data_utils import *
reader = dataPreprocessor(path_to_record_name, path_to_noisy, path_to_clean, use_waveform=True)
reader.write_tfrecord()

Training phase

python main.py stage1

Testing phase

In main.py

change test_path and test_list

python main.py test