Speech Denoising with Deep Feature Losses (arXiv, sound examples)
This is a Tensorflow implementation of the following research paper Speech Denoising Convolutional Neural Network trained with Deep Feature Losses.
If you use the code for research, please cite the paper: François G. Germain, Qifeng Chen, and Vladlen Koltun. Speech Denoising with Deep Feature Losses. arXiv:1806.10522. 2018.
The source code is published under the MIT license. See LICENSE for details. In general, you can use the code for any purpose with proper attribution. If you do something interesting with the code, we'll be happy to know. Feel free to contact us.
Clone the repo into your local machine by making use of Git commands. You must have a python2 environment setup before proceeding any further. It is recommended to create a virtual environment of python2.7. You can create a virtual environment using the package 'virtualenv'. Install the package by 'pip install virtualenv'.
Now setup a virtualenv with the help of following command: 'virtualenv venv --python=python2.7'
Activate the venv with the following command: 'source venv/bin/activate'
Required python libraries: Tensorflow with GPU support (>=1.4) + Scipy (>=1.1) + Numpy (>=1.14) + Tqdm (>=4.0.0). To install in your python2.7 virtualenv, run
pip install -r requirements.txt
Warning: Make sure your libraries (Cuda, Cudnn,...) are compatible with the tensorflow version you're using or the code will not run.
Run the script 'app.py' with the help of the command 'python app.py' on your CLI [NOTE: python here refers to python2 assuming you are operating inside the venv].
This will output an IP address on your terminal screen, which you can open in a supported browser (tested on Safari). The IP address will prompt you to upload a noisy audio file you want to denoise. Follow the given instructions to proceed further and soon you will have your desired denoised samples.