Rnnoise, neural network noise canceling - AudioBridge

[mirkobrankovic]

Based upon rnnoise library, and only as a poc, to see if it makes sense to use it or not.
Instalation instructions are updated in README.md and also audiobridge usage.

rnnoise_complexity room parameter is actually how often frames should go trough rnnoise.
0 - disabled
1 - every second frame will be denoised
2 - 2 frames will be denoised, third skipped
3 - every frame will be denoised
This is an option since denoising will decrees audio quality (add a bit robotic flavor or even sometimes cut end of words, recognizing them as noise)

The rnnoise readme states It operates on RAW 16-bit (machine endian) mono PCM files sampled at 48 kHz and from code example I see it does read from a file into frames and process them one by one. I only managed to get clear audio on 16000 sampling rate room.

It would be nice to get some feedback on how good it is, or even better if it can be made better in any way, since my understanding of sampling is very limited 👍 )

[lminiero]

Sorry, but I'm not going to consider major changes like this, especially when they add new dependencies, until the multistream branch has been merged.

[mirkobrankovic]

aah make sense, thanks :)

[lminiero]

Hey @mirkobrankovic, I was thinking about noise suppression and this old PR of yours came back to mind: would you mind creating a new PR for this effort when you have some time? At the time we could not deal with it because multistream hadn't been merged yet, but now that it has I'm curious to see what kind of impact it may have on audio quality in the AudioBridge. It's interesting that rrnoise can only work at 16000 as a sampling rate, for instance. Thanks!