# speech-enhancement-WGAN

**Repository Path**: immortal_soul_of_sword/speech-enhancement-WGAN

## Basic Information

- **Project Name**: speech-enhancement-WGAN
- **Description**: speech enhancement GAN on waveform/log-power-spectrum data using Improved WGAN 
- **Primary Language**: Unknown
- **License**: Not specified
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 2
- **Forks**: 0
- **Created**: 2019-12-25
- **Last Updated**: 2023-07-25

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# 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)__

```python
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__

```bash
python main.py stage1
```
__Testing phase__

In main.py

change test_path and test_list

```bash
python main.py test
```