Discord: Denoise Your Friends

Instructions for how to denoise your friend's microphones on Discord.

Motivation

On Windows, Discord provides noise suppression via Krisp.ai to denoise your microphone.

Awesome!

... but what if I need to denoise my friends' microphones?

This can happen for various reasons. E.g.

• Your friends refuse to turn on noise suppression.
• Your friends are on a platform where Krisp.ai is not available.
• You've finally reached your limit.

It'd be awesome if we could do this easily from within discord itself, but we can't (yet?).

Ok, so what's the plan?

Insert Xiph.org's RNNoise Suppressor into our audio pipeline, but only for voice applications. This algorithm has already been conveniently packaged as an audio plugin for Windows and Linux by Werman over at https://github.com/werman/noise-suppression-for-voice

Some Questions...

Q: But what if I just want to denoise my own my own microphone, but I'm on Linux?

• A: Follow the instructions for Pulse Audio / LADSPA here:

  https://github.com/werman/noise-suppression-for-voice

Q: Will this affect my music/movies/games/etc?

• A: No. It only affects applications that you explicitly route to the noise suppressor.

Q: Can I use this AND denoise my mic also?

• A: Yes.

Status

These instructions should be considered beta. They are mostly my notes from having done this before, but could use further testing.

Instructions for Ubuntu Linux

Overview: We need to create a virtual audio sink to send the Discord output to, then run the above listed noise suppression on it.

1. Install prerequisites

   $ sudo apt-get update
   $ sudo apt-get install -y git build-essential cmake
   

2. Download and compile the noise-suppression-for-voice plugin like so:

   $ git clone https://github.com/werman/noise-suppression-for-voice.git
   $ cd noise-suppression-for-voice
   $ mkdir build
   $ cd build
   $ cmake -DCMAKE_BUILD_TYPE=Release ..
   $ make -j
   $ sudo mkdir -p /opt/noise-suppression/lib
   $ sudo cp bin/ladspa/librnnoise_ladspa.so /opt/noise-suppression/lib
   

3. The following commands will setup a virtual pulseaudio device for you. However, they need to be executed every time you login, so we will put them in ~/.config/pulse/default.pa

   If you just want to test things out, you can use the one-off command-line version below. Otherwise, you can use the default.pa file shown in the second code block below.

   One-off command-line version

   $ pacmd load-module module-ladspa-sink sink_name=discord_denoise_input label=noise_suppressor_stereo plugin=/opt/noise-suppression/lib/librnnoise_ladspa.so control=50
   
   $ pacmd load-module module-null-sink sink_name=discord_input sink_properties=device.description=Discord_Denoiser_Input
   
   $ pacmd load-module module-loopback source=discord_input.monitor sink=discord_denoise_input source_dont_move=true sink_dont_move=true latency_msec=1
   

   Automatic Login Version for ~/.config/pulse/default.pa

   .include /etc/pulse/default.pa
   
   # final virtual output device; processes the noise suppression
   load-module module-ladspa-sink sink_name=discord_denoise_input label=noise_suppressor_stereo plugin=/opt/noise-suppression/lib/librnnoise_ladspa.so control=50
   
   # virtual device that we connect discord output to
   load-module module-null-sink sink_name=discord_input sink_properties=device.description=Discord_Denoiser_Input
   
   # discord output gets forwarded to the noise suppressor
   load-module module-loopback source=discord_input.monitor sink=discord_denoise_input source_dont_move=true sink_dont_move=true latency_msec=1
   

4. Configure Discord:

   • Go into Discord -> Settings -> Voice Settings -> Output Device and select Discord_Denoiser_Input as the Output Device

5. (Optional) Additional Configuration:

   • Start Pulse Audio Volume Control, pavucontrol, and you should see two new virtual devices: Discord_Denoiser_Input and also LADSPA Plugin Noise Suppressor for Voice (Stereo). Just make sure both are unmuted and at 100% volume. You can adjust these if necessary.

6. Test it with your friends

   • You can change Discord's output between your Default device and the Discord_Denoiser_Input device to hear the difference.

   • You can make the filter more or less aggressive by tweaking the control=50 value in default.pa. This value controls the probability threshold that a sound is voice or not. If the sound is lower than this value, the noise suppressor returns silence. I find that 50 works pretty well for my friend's loud keyboard noises. But feel free to tweak it to your liking.

7. Troubleshooting.

   • I noticed on my machine that sometimes I get crackling/popping on non-denoised channels when this is enabled. This plugin seems to work best on my machine when applying the tsched=0 fix described here:

   https://wiki.archlinux.org/title/PulseAudio/Troubleshooting#Glitches,_skips_or_crackling

   That page has a lot of useful troubleshooting ideas if you experience issues.

Windows

We need a couple pieces of software to accomplish this:

• The free version of Krisp.ai, which we use to create a virtual audio device. Download it from Krisp. There's probably other software you can use to create a virtual audio device. I've only tested with Krisp.ai. We will disable the Krisp algorithms and just use the denoiser mentioned above.

• A Windows build of the noise suppressor above, which you can download from the releases page: https://github.com/werman/noise-suppression-for-voice/releases

• EqualizerAPO

Instructions:

1. Download and install Krisp. Krisp will create a virtual audio device for you automatically. The Krisp control panel lives in the Windows System Tray. Open it, and disable the Krisp algorithms. You don't need to go through any of the steps they recommend to setup voice suppression.

2. Unpack librnnoise_vst.dll from windows_rnnoise_bin_x64.zip to C:\Program Files\Common Files\VST2\librnnoise_vst.dll. You may need to create the parent folder if it doesn't exist. That's a standard location for VST's, but you can put it somewhere else if you like, e.g. C:\VST2.

3. Install EqualizerAPO

4. Start the EqualizerAPO Configurator.exe (found in C:\Program Files\EqualizerAPO\) and check the checkbox for Krisp Speaker to install the APO on the virtual Krisp device. You'll probably need to reboot after doing this.

5. Start the EqualizerAPO Editor.exe (found in the same location). You'll need to do a couple things there:

   1. Set the Selected Device to Krisp Speaker
   2. Use the green + Button to create a Plugins -> VST Plugin
   3. Load the VST .dll file you unpacked earlier.
   4. Make sure the Power/On Icon is enabled for both devices
   5. Click the Save Button on the toolbar to save your changes.

   NOTE: DONT click the Open Panel panel button. Usually, this would open the VST2 configuration interface, but this particular VST doesn't actually have a GUI, so APO might crash if you do that.

   The resulting configuration should look like this:

   

6. Configure Discord:

   • Go into Discord -> Settings -> Voice Settings -> Output Device and select Krisp Speaker as the Output Device

7. Test it.

   • You can toggle the noise suppressor on and off with the power button that's on the VST plugin area in the EqualizerAPO Editor.exe

Version History

• 2022-07-16: Updated with tsched=0 fix
• 2022-07-11: Add note about pulseaudio priority
• 2022-07-10: Fix minor typos
• 2022-07-09: Initial Version by ahrbe1