Note

This repository is intended as a reference for ideal ratio mask dnn implementation

It is based on my university dissertation. Please send me an email if you would like to know more about it.

Python Version

Python 3 (Python 2 also works with some modifications, search TODO in the repository)

Installing dependencies

run the following command from the project root directory pip install -r requirements.txt

PESQ software: get the PESQ software from http://www.itu.int/rec/T-REC-P.862-200511-I!Amd2/en and run the following command in the software's source directory: gcc -o pesq *.c Place the binary file in any $PATH directory

Acquire Full Dataset

Download LibriSpeech dataset from http://www.openslr.org/12/ and place the uncompressed files to the appropriate directory

Download the DEMAND dataset from https://datashare.is.ed.ac.uk/handle/10283/2791 and place the uncompressed files to the appropriate directory

Generating Training Dataset

cd preprocessing
python generate_dataset.py

Train a Neural Network (Example)

cd train
python residual_train.py

Test a Neural Network (Example)

cd test
python test_snr_performance.py