Tensorflow 2.0 implementation of the paper A Fully Convolutional Neural Network for Speech Enhancement
This code is a standalone extension of [this code] (https://raw.githubusercontent.com/daitan-innovation/cnn-audio-denoiser/) Which in turn is based on this blog post: Practical Deep Learning Audio Denoising
Although a partial part of the dataset is said to be available on the above github repo, it isn't there. Therefore you need to download the two datasets below.
You can then use create_dataset.py
script to create the TFRecord files.
You will then need to update the paths of these datasets in default_config.py
.
Use train.py to train an audio denoiser
Use test.py to test the audio denoiser with one track
You need to install tensorflow and keras and some other bits and pieces. It would be best to follow the guidance on installing TF on it's website (i.e. use within a virtual environment)