search

nay0648 / unified2021

A UNIFIED SPEECH ENHANCEMENT FRONT-END FOR ONLINE DEREVERBERATION, ACOUSTIC ECHO CANCELLATION, AND SOURCE SEPARATION
110 stars 55 forks source link

solve gev problem and the scaling ambiguity #3

Open meadow163 opened 2 years ago

meadow163 commented 2 years ago

Hi nay0648,

Thank for your amazing work and it really helps me a lot in my research. But I met a problem when I run your BSS code using a four channels data, could you please tell me how to change your BSS code to support multi-channel processing?

Best Regards.

Thanks in advanced!

nay0648 commented 2 years ago

Hello,

Thanks for watching! My provided code on github is only supported for 2 channel BSS, the demixing matrix can be acquired analytically by solving the 2 x 2 generalized eigendecomposition problem.

However, for channels more than 2, there is no closed-form solution, instead, the matrix inversion lemma can be used to solve the problem iteratively. Please refer to the paper for the detailed theory (section 4.2). If the computational complexity is not considered, the easier way is to perform matrix inversion directly by matlab’s inv function (Algorithm 1 in the paper).

Taniguchi, Toru, et al. "An auxiliary-function approach to online independent vector analysis for real-time blind source separation." 2014 4th Joint Workshop on Hands-free Speech Communication and Microphone Arrays (HSCMA). IEEE, 2014.

Best Regards.

At 2022-08-08 11:53:35, "meadow163" @.***> wrote:

Hi nay0648,

Thank for your amazing work and it really helps me a lot in my research. But I met a problem when I run your BSS code using a four channels data, could you please tell me how to change your BSS code to support multi-channel processing?

Best Regards.

Thanks in advanced!

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you are subscribed to this thread.Message ID: @.***>