DDAE speech enhancement

Hyper-parameters are not tuned to optimal

Prerequisites:

• Keras 1.2
• Tensorflow 1.x as backend
• h5py
• librosa
• scipy

Getting Started:

Extract spectrogram features:

python spectrum.py data.h5 list_noisy list_clean

Train DDAE using Keras:

python train_DNN.py data.h5

Enhance test wave files using trained model:

python test_gen_spec.py model.hdf5 list_noisy