NoiseTorch is an easy to use open source application for Linux with PulseAudio. It creates a virtual microphone that suppresses noise, in any application. Use whichever conferencing or VOIP application you like and simply select the NoiseTorch Virtual Microphone as input to torch the sound of your mechanical keyboard, computer fans, trains and the likes.
If noisetorch doesn't start after installation, you may also have to make sure that `.local/bin` is in your PATH. On most distributions e.g. Ubuntu, this should be the case by default. If it's not, make sure to append
Please do not use them. Some are known to mess with configuration files and will break things even if you did install a NoiseTorch package from an official source later on. If you had ever installed NoiseTorch from a non-official source, make sure to *completely* remove every trace, including deleting your `~/.config/noisetorch` directory, before reporting an issue.
If you have been maintaining a third party package for a while and would like it to be blessed as supported, please send me an e-mail.
Select the microphone you want to denoise, and click "Load NoiseTorch", NoiseTorch will create a virtual microphone called "NoiseTorch Microphone" that you can select in any application.
When you're done using it, simply click "Unload NoiseTorch" to remove it again, until you need it next time.
The slider "Voice Activation Threshold" under settings, allows you to choose how strict NoiseTorch should be in only allowing your microphone to send sounds when it detects voice.. Generally you want this up as high as possible. With a decent microphone, you can turn this to the maximum of 95%. If you cut out during talking, slowly lower this strictness until you find a value that works for you.
If you set this to 0%, NoiseTorch will still dampen noise, but not deactivate your microphone if it doesn't detect voice.
Once NoiseTorch has been loaded, feel free to close the window, the virtual microphone will continue working until you explicitly unload it. The NoiseTorch process is not required anymore once it has been loaded.
NoiseTorch may introduce a small amount of latency. The amount of inherent latency introduced by noise supression is 10ms, this is very low and should not be a problem. Additionally PulseAudio currently introduces a variable amount of latency that depends on your system. Lowering this latency [requires a change in PulseAudio](https://gitlab.freedesktop.org/pulseaudio/pulseaudio/-/issues/120).
* [@werman](https://github.com/werman/)'s [LADSPA/VST wrapper](https://github.com/werman/noise-suppression-for-voice/) allowing us to load RNNoise into PulseAudio.
* [Salee Design](https://www.salleedesign.com) (info@salleedesign.com)'s Microphone Icon under [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/)
