/code_2020ICASSP_iss

Code to reproduce the experiments in the paper "Fast and stable blind source separation with rank-1 updates" presented at ICASSP 2020.

Primary LanguagePython

Fast and stable blind source separation with rank-1 updates

Abstract

We propose a new algorithm for the blind source separation of acoustic sources. This algorithm is an alternative to the popular auxiliary function based independent vector analysis using iterative projection (AuxIVA-IP). It optimizes the same cost function, but instead of alternate updates of the rows of the demixing matrix, we propose a sequence of rank-1 updates. Remarkably, and unlike the previous method, the resulting updates do not require matrix inversion. Moreover, their computational complexity is quadratic in the number of microphones, rather than cubic in AuxIVA-IP. In addition, we show that the new method can be derived as alternate updates of the steering vectors of sources. Accordingly, we name the method iterative source steering (AuxIVA-ISS). Finally, we confirm in simulated experiments that the proposed algorithm separates sources just as well as AuxIVA-IP, at a lower computational cost.

Authors

Install

We use anaconda to setup the Python environment.

git clone --recursive <url>
cd piva
conda env create -f environment.yml
conda activate piva
python setup.py build_ext --inplace
cd ..

Run Experiments

The two experiments presented in the paper can be run by the following steps. This produces two files ./experiment_metrics_speed_results.json ./experiment_speed_11_17_results.json that are later used to produce the plots.

conda activate piva

# Run the simulations
python ./experiment_metrics_speed.py
python ./experiment_speed_11_17.py

# Plot the results
python ./make_figures.py ./experiment_metrics_speed_results.json ./experiment_speed_11_17_results.json

The two simulation output data files produced for the figures in the paper were kept in the sim_results folder.

License

The code in this repository is released under the MIT license.