RayanAAY-ops/RE-Deep-Recurrent-NMF-for-Speech-Separation-by-Unfolding-Iterative-Thresholding

In this code, we propose a different implementation fo the original paper https://arxiv.org/abs/1709.07124 under PyTorch. This architecture is constructed by unfolding the iterations of a sequential iterative soft-thresholding algorithm (ISTA) that solves the optimization problem for sparse nonnegative matrix factorization (NMF) of spectrograms. We name this network architecture deep recurrent NMF (DR-NMF)

Deep-Unfolded-NMF-for-Speech-Source-Separation

This project is an implementation of the paper "Deep Recurrent NMF for Speech Separation by Unfolding Iterative Thresholding" of the authors : Scott Wisdom, Thomas Powers, James Pitton and Les Atlas. The original paper code has two major constraints :

• It was developped using both Python and uses API calls to Matlab.
• The code is highly dependent on the private dataset CHiME2.

In this project we propose a full implementation in Python 3 (No Matlab) that can be used with wav files (more general).

We have updated the original algorithm in different ways to increase the temporal consistency of the audio data, the description in details is given in the final report. This approach allows a better generelization and interpretation of what the model is performing compared to SOTA deep learning.

Here we can observe the loss evolution and the increases of SDR in both speech and music.

@article{DBLP:journals/corr/abs-1709-07124, author = {Scott Wisdom and Thomas Powers and James W. Pitton and Les Atlas}, title = {Deep Recurrent {NMF} for Speech Separation by Unfolding Iterative Thresholding}, journal = {CoRR}, volume = {abs/1709.07124}, year = {2017}, url = {http://arxiv.org/abs/1709.07124}, eprinttype = {arXiv}, eprint = {1709.07124}, timestamp = {Mon, 13 Aug 2018 16:48:55 +0200}, biburl = {https://dblp.org/rec/journals/corr/abs-1709-07124.bib}, bibsource = {dblp computer science bibliography, https://dblp.org} }