Bayesian anisotropic Gaussian model for audio source separation

Paul Magron and Tuomas Virtanen.
Proc. of the IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), April 2018.
[Paper] [Slides]

On this page, we provide several audio excerpts that illustrate the source separation experiments presented in the paper. The songs are extracted from the Demixing Secrets Database (DSD100).

We particularly note that the bass track is neater when estimated with our method compared to the others, and the drum track contains less artifacts.

Song: "One Minute Smile" by Actions

Mixture: 


Bass Drum Other Vocals
Original sources
Wiener filtering
Consistent Wiener filtering
Anisotropic Wiener filtering
Proposed estimator

Song: "I'm alright" by Angels in Amplifiers

Mixture:


Bass Drum Other Vocals
Original sources
Wiener filtering
Consistent Wiener filtering
Anisotropic Wiener filtering
Proposed estimator