Beamforming and dereverbaration

Post Reply
Suf
Posts: 2
Joined: Thu Feb 20, 2020 1:30 pm

Beamforming and dereverbaration

Post by Suf » Thu Feb 20, 2020 1:37 pm

Hi!

Thank you for this brilliant, open-source, piece of software.

I have a few questions which I'm hoping to get some help and understand.

1. Following your live Gain example, I'm trying to connect to JACK, apply some gain, but instead of playing this through the speakers, I'd like to record this on a file. Is it possible to use to iolibs at once? Any tips or pointers to do so?

2. Using your binaural beamforming example, I'm trying to tweak it to work for 8 mics instead of just 4. What would I need to do in that case? Would I need to generate vectors and place them on a text file as the example does?

3. I'm trying to apply some dereverberation on a music file recorded from a microphone. Would the coherence example help in this case?

Thank you in advance!

hendrikkayser
Posts: 3
Joined: Fri May 10, 2019 7:58 am

Re: Beamforming and dereverbaration

Post by hendrikkayser » Thu Feb 20, 2020 3:27 pm

Hello!

re 1.: It is not possible to use two iolibs, but you can add the plugin 'wavrec' to your mhachain to record audio.

re 2.: You'll find a description of the structure required for the file containing the filter coefficients in the documentation of the acSteer plugin in the plugin manual (www.openmha.org/docs/openMHA_plugins.pdf).

re3.: The coherence filter computes coherence values between two audio input channels. If two mics were used for the recording, give it a try! An play a little with the parameters...

I hope this helps!

Suf
Posts: 2
Joined: Thu Feb 20, 2020 1:30 pm

Re: Beamforming and dereverbaration

Post by Suf » Thu Feb 20, 2020 5:16 pm

Brilliant, thank you!

Is there a plugin for openMHA or some kind of python script that can help me create this file needed by acSteer?

hendrikkayser
Posts: 3
Joined: Fri May 10, 2019 7:58 am

Re: Beamforming and dereverbaration

Post by hendrikkayser » Thu Feb 27, 2020 12:57 pm

Unfortunately, we don't have a proper script at hand at the moment and I'm aware that we need to provide one in the future. The computation of the MVDR beamformer from head-related transfer functions is described in this contribution presented at the ITG Conference on Speech Communication 2016:

www.sigproc.uni-oldenburg.de/download/p ... 016_17.pdf

Post Reply