SemiBlindICA working examples?

[ax]

Hello HARK support,

If exists, could you share any working example which demonstrates SemiBlindICA?

I need AEC on my mic array (sure it has a reference channel) but I cannot make SemiBlindICA works properly.
Network works without any error, but the localized/separated sound is not what I need:
– sometimes self-noise is localized while user’s speech is filtered
– sometimes user’s speech is localized, but on my ear, the separated sound contains loud self-noise

Note that the “normal” (without self-noize) localization/separation works fine with the same microphone.

Can you point me if any topical mis-configuration regarding SemiBlindICA? or if possible can you share any working examples?

Thanks,

[Yuka Goto]

Dear ax,

Unfortunately, we do not have a sample network which demonstrates SemiBlindICA at hand, but we suspect that the TAP_LOWFREQ value may be causing the problem.
This value needs to be set larger when the environment contains a long reverberation as described in the document: https://www.hark.jp/document/hark-document-en/subsec-SemiBlindICA.html

A spacious environment like concert halls/churches or setting a microphone array and a speaker apart from each other cause “a long reverberation.”

If these are not applicable, please give us more details of your working environment and parameter vaules you set on SemiBlindICA.

Sincerely,

HARK Support Team

[ax]

> but we suspect that the TAP_LOWFREQ value may be causing the problem.
Thanks, by setting both TAP_HIGHFREQ and TAP_LOWFREQ to zero, my network separates speeches as expected! (still some noises are separated as sep files anyway)

Thank you for your help!

[Author]

Posts