/DeepDenoisingAutoencoder

Tensorflow implementation for Speech Enhancement (DDAE)

Primary LanguagePythonMIT LicenseMIT

Deep Denoising Autoencoder (DDAE) for Speech Enhancement

Tensorflow implementation of Speech Enhancement Based on Deep Denoising Autoencoder

Getting Started

Clone This repository to your local machine and run create_dir.sh first.

Prerequisites

  • python 3.5
  • tensorflow-gpu 1.8.0
  • scikit-learn 0.19.1
  • scipy 1.1.0
  • h5py 2.7.1
  • librosa 0.5.1
  • numpy 1.14.3
  • tqdm 4.23.2

Getting Started

  1. Download free dataset from VoxForge for clean data. Here I would recommed download cmu_us_awb_arctic.tgz
  2. Unzip clean dataset to /DeepDenoisingAutoencoder/data/raw/clean/
  3. Download free dataset from ESC-50 for noise data.
  4. Move ESC-50-master/audio to /DeepDenoisingAutoencoder/data/raw/noise/
  5. Set parameters in python/main.py
  6. Run python/main.py

Result

Spectrogram on Test data

Deployment

You can read many comments inside all .py files.

Authors

Yu-Ding Lu - Linkedin

License

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

Acknowledgments

  • Bio-ASP lab - CITI - Academia Sinica