Hacker News Reader

Comment by kevincox | Hacker News Reader

It isn't quite the same, but RNNoise is a neural-network based noise reduction aimed at dampening anything that isn't human voice. It works fantastic to me and almost always completely mutes my keyboard. (Sometimes it comes though a bit if I am talking at the same time.)

This is easy integrated into Pulseaudio using PulseEffects (https://github.com/wwmm/pulseeffects).

(The PulseEffects support is new, I used to use NoiseTorch (https://github.com/lawl/NoiseTorch) which is more dedicated but has a weird UI and some other downsides)

View on HN

> and some other downsides

Can you elaborate? (maintainer here, always interested in feedback)

For me, I think having an option to start when I login would be great (similar to NVIDIA Broadcast on Windows).

This would solve my biggest issue that when turning on and switching my input to it, other programs sometimes need to be restarted to pick up the microphone change.

Otherwise, keep up the great work!

I can testify that these RNNoise-based methods work quite well at removing background noise, including keyboard clacking, even while talking and typing at the same time.

I've used this in Windows and it works pretty good. The filter isn't quite as good as RTXVoice and the installation can be... tricky[1], but it's less fiddly once you get it working.

[1] https://medium.com/@bssankaran/free-and-open-source-software...

RNNoise is also available as a native voice filter in OBS and it works very well there.

For anyone going down this path, it's relatively easy to kinda use OBS as your audio pipeline for other programs. (Easy in the "for people on HN" sense, not easy in the "my mom would do this" sense.) Only caveat is that you need to leave OBS running in the background doing its thing whenever you need your microphone.

(Instructions for Windows. The OBS stuff would be the same, you'll just need to figure out your own solution for virtual audio devices.)

1. You'll need a virtual audio device. So grab Virtual Audio Cable[1] or VB-Audio Cable[2]. Both have a free version that let you set up a single device.

2. In OBS, add your microphone as an "Audio Input Capture". Right click on it, select Filters, and add a "Noise Suppression". (For future playing, you may want to come back here and add a "Noise Gate" as well. Basically "if the audio is too quiet, mute the input completely.)

3. Now go to File -> Settings, Audio, and under "Advanced" select the virtual audio cable as your monitoring device. Hit OK.

4. In the Audio Mixer panel at the bottom middle of your window, select any of the gear icons and hit "Advanced Audio Properties". In the row for your microphone input, change "Audio Monitoring" to "Monitor and Output".

5. You should now be getting your de-noised/gated/etc microphone coming in to the virtual audio device. Now in your voice chat program, just change the microphone input to the virtual audio device.

I've got a relatively quiet workspace, but I still live with three people and two dogs so the ambient noise is there. This cuts most of my dog's sounds (even if she's playing and creating a ruckus), ambient noise from people upstairs, my keyboard, etc. I'm stuck in meetings a good chunk of my day so I try to avoid having to wear headphones (just gets tiresome), and the noise gating and some strategic microphone positioning works perfect for keeping the meeting audio from my speakers from feeding back in. (Along with having all the meeting audio running through a very aggressive compressor to normalize everyone's volume into a very tight range.)

OBS also includes a virtual camera option, so I use that to crop out a tighter shot from my fairly wide angle camera so people can't see as much of the mess around the room and do some colour adjustment because it's a cheap camera and looks kinda garbage otherwise.

[1] https://vac.muzychenko.net/en/ [2] https://vb-audio.com/Cable/

Doing that on a Mac is...more complicated. There's only one company (Rogue Amoeba) that does virtual audio devices (https://rogueamoeba.com/loopback/) and it's $100.

Incidentally I purchased an Elgato Wave:3 Mic (https://www.elgato.com/en/wave-3) because it includes a free virtual audio device for Mac, which piqued my curiosity. It turns out that Rogue Ameoba provides the extension for virtual audio devices, so it's a good value!

Another option for both Win/Mac is the Blue Yeti X (https://www.bluemic.com/en-us/products/yeti-x/), which has Blue Voice to denoise/noise gate/equalize audio without additional drivers aside from the base software (I use a headset which supports Blue Voice and it's ok, but def not as good as RNNVoice)

Thanks for this extremely detailed write up. I also appreciated the tips on compressing the dynamic range of audio.

I've just used a little script that sets up https://github.com/werman/noise-suppression-for-voice and it's worked wonderfully, but PulseEffects and NoiseTorch seem cool!

Seems like this would be great for multiplayer gaming. Anyone know how well PulseAudio works on windows these days?

Any recommendation for mac?

Try Unclack? https://news.ycombinator.com/item?id=25644828

I presume he/she meant - Is there a way to get RNNoise working on Mac?

Here:

https://stackoverflow.com/a/59808919/1660678

Then you still get a FAILED: meson-install Permission denied: '/Library/Audio/Plug-Ins/LV2' error, so change INSTALL_DIR_MAC in install.sh to~/Library/Audio/Plug-Ins/LV2 and install via the github page

krisp.ai is good but $3 a month

For what it's worth, I tried it and it was... not great?

Versus piping my audio through OBS as a way of using RNNoise and doing noise gating, OBS retains the original audio quality quite well and does a really good job of removing clicks/clacks/etc.

Krisp made my voice sound like I was talking through two tin cans and a string, and, at least in my environment, did not do a very good job of actually removing any of the background sounds.

I've also tried it in Discord (as it's integrated directly into Discord) and kept running into a problem of it deciding my own voice was noise and just cutting me out mid word.

They have a free tier (gives you 120min/week of noise cancelling), so anyone looking to try it definitely worth using that first. I'd suggest turning on audio monitoring in some headphones and listening to yourself a bit though and deciding if the quality is acceptable to you.

I've only had excellent experience with Krisp, but I have a semi-pro USB mic I use so perhaps it makes all the difference. Sounds great also with a macbook built-in mic.

That said, I'm interested to see open source alternatives as I'm interested to see how they work.

I've tried it with both a $20 off-brand shotgun mic as well as a Blue Yeti. Not much luck with either. RNNoise did vastly better and the work getting it set up in my audio pipeline certainly wasn't worth $5/mo to avoid.

And honestly I could've even lived with that except for the severe degradation in audio quality.

Reply on news.ycombinator.com