This repository contains implementations of various source separation algorithms in Python. Presently, all algorithms are based on non-negative matrix factorization and its variants. More implementations using other techniques will follow. While the presentation of the NMF algrithms and its variants is work done as a part of the course EE4902, this also contributed quite significantly to a course project completed as a part of the course EE5120 at IIT Madras.
All code is written in Python (and functional in version 3.6.8) and uses the libraries listed in environment.yml. This assumes that you have miniconda installed. If you do not, get it here.
Create a conda environment in which all the code can be run using the following command in the terminal:
conda env create -f environment.yml
This will create an environment named source_separation (name can be changed by changing the value of "name" in the environment.yml file). Activate the environment using
conda activate source_separation
and run the desired script as described.
Monoaural_source_separation_using_Non-negative_matrix_factorization: Code for source separation based on the technique described in the paper Monaural Sound Source Separation by NonnegativeMatrix Factorization With Temporal Continuity and Sparseness Criteria.
sourcesep: Modules for all implemented source separation algorithms
tools: Modules for helper functions