NewsWorksSoftwareTextBioContact
background image

Denoising of ambisonic field recordings

January 11, 2016

Screen_shot_2016-01-11_at_15

I’ve been experimenting lately with what might be the best way to do noise reduction (denoising) of ambisonic field recordings. In some of the recordings from very quiet locations preramp noise is noticeable. I’m not interested in reducing any of the background noise that is part of the recorded ambience itself, I just want to reduce the amount of hiss introduced in the process of recording.

The recordings are done using a SoundField SPS200 mic and a SoundDevice 788T recorder. The resulting recordings are A-format, and I use the SoundField SPS200 plugin to convert the A-format recordings to B-format. Denoising is done using iZotope RX. RX4 is not able to deal with multichannel files, so I have to split the 4-channel recording into two stereo tracks, do noise cancellation on them, and then merge back into four channels again. However there are at least three different approaches to this:

  1. Treat the original A-format recordings, and afterwards convert the processed files to B-format using the SPS200 plug.
  2. Treat the converted B-format recordings.
  3. Decode the B-format files to A-format using the BtoA decoder from Ambisonic Toolkit for Reaper, remove noise, merge, and then re-encode to B-format using the ATK AtoB encoder.

I have done recordings of the SoundDevice at various input levels with no mic connected. The resulting hiss is very similar to the noise that I want to remove, and I use these recordings to train the Denoiser. The resulting curves are stored as a set of presets.

After having experimented over several days, and a lot of useful input from the sursound list, it seems to me that processing of the B-format signals alter some of the spatial information. The resulting spatial image appears to me to be more unstable, and less spatially articulated. If the Denoiser algorithm do alter phase I suspect that it might also impact the original A-format recordings negatively. So I am ending up with the last option above.

One benefit of this is that I can do it all in Reaper, and I don’t have to manually split and merge sound files. It is still a laborious process: At my 4 year old laptop rendering is happening at 1.1 x realtime speed, so there is a lot of waiting.

This blog post was done while waiting for a rendering that took close to 50 minutes to complete. Duh!