/speech-enhancement-WGAN

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

Primary LanguagePython

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