tool status + guide
NoiseTorch in 2026: dormant, still usable, mostly replaceable
NoiseTorch is the most famous name in Linux noise suppression and the least current advice on the internet about it. Every aging listicle still opens with it. So let's do what those pages don't and start with the repository facts, checked today, not in 2021.
- 1v0.12.2 — the latest release
- 2a commit on the default branch
- 3the last commit we could find
- 4checked: still no newer release; 70 open issues
What NoiseTorch is (and why everyone loved it)
NoiseTorch is a small Go application that puts RNNoise behind the single friendliest interface this niche has ever produced: pick your mic, click Load NoiseTorch, and a new input device called "NoiseTorch Microphone" appears system-wide. Select it in any app and that app hears denoised audio. One slider (voice activation threshold), one button, done. It talks the PulseAudio API, which also makes it work on PipeWire systems through the pipewire-pulse compatibility layer — the project's own README says "PulseAudio or PipeWire," and that's accurate in our experience with the compat route.
That UX is why it went viral in the work-from-home era, and credit where due: for PulseAudio desktops in 2020–2022, it was simply the right answer.
The 2022 security scare, briefly
In 2022 the original developer reported a suspected compromise related to the project's infrastructure, and development as users knew it stopped. The community response was genuinely good: volunteers audited the codebase, the README today states plainly that no malicious code was found, and the project continued under community stewardship as "NoiseTorch-ng," with updates distributed through GitHub releases. We wrote up the whole saga — dates, handoff, aftermath — in a separate timeline post, because it's the clearest case study we know of what "maintained" does and doesn't mean in open source.
The upshot for you: the code is considered clean. The concern in 2026 isn't malware — it's bitrot. A tool that manipulates your audio server's internals, frozen at a 2022 snapshot, meets a PipeWire that has moved through years of releases since. It still works today on mainstream setups. Each passing release makes that a little less guaranteed, and 70 open issues are aging in the queue with no one obligated to answer them.
Honest recommendation matrix
| Your situation | Our call |
|---|---|
| Current distro, PipeWire (the 2026 default) | Skip NoiseTorch. EasyEffects does the same job, is released monthly-ish, and is in your official repos. |
| Older box genuinely still on PulseAudio | NoiseTorch remains a fine, low-effort choice. This is its last legitimate niche and we won't pretend otherwise. |
| You want zero GUI anywhere | Use the filter-chain instead — actively maintained plugin, pure config. |
| You loved the one-toggle UX | EasyEffects with an autoloading preset is one-time setup, zero-toggle thereafter — effectively even less UX. |
Setup, if you've decided it's your niche
NoiseTorch isn't in major official repos, so installation is manual or via AUR:
~$ # Arch-family, via AUR helper (AUR, not official repos):
~$ paru -S noisetorch
~$ # Everyone else: download the release tarball from
~$ # github.com/noisetorch/NoiseTorch/releases (v0.12.2),
~$ # verify the checksum they publish, then:
~$ tar -C ~ -xzf NoiseTorch_x64_v0.12.2.tgz
~$ ~/.local/bin/noisetorchIn the app: select your physical microphone, leave the voice activation threshold near its default to start (raise it if noise sneaks through in silences, lower it if your quiet syllables get clipped away), and click Load NoiseTorch. Then pick "NoiseTorch Microphone" as the input in your calling or recording app. To undo everything, click Unload NoiseTorch — it removes the virtual device cleanly.
Two operational notes from real use. First, load it before joining a call; apps enumerate devices at startup and some never refresh the list. Second, don't run it alongside another suppressor — see the shared-model warning — and especially don't point EasyEffects at NoiseTorch's virtual mic, which is a warble machine.
Check the dormancy yourself (it takes two minutes)
Don't take our word for any of this — the whole point of open source is that the evidence is public. These are the exact four checks behind the status box above, and they work on any project you're considering:
- Latest release, not latest commit. The releases page is what users actually install. A repo can look busy with typo fixes while shipping nothing for years — which is precisely NoiseTorch's shape.
- Last commit on the default branch. Here: January 2025, and the one before it was April 2024. Two commits in two years is a maintenance pulse, not a heartbeat.
- Is it in official distro repos? Distro maintainers are a free early-warning system: they drop packages that stop building or stop being supported. AUR-only, with nothing in Debian, tells you what Debian concluded.
- Are issues being answered? Seventy open, with the recent ones unanswered, means you are the support team if something breaks.
None of those four checks is about audio quality, and that's deliberate. NoiseTorch's audio is fine — it's the same RNNoise everything else runs. The risk you're pricing is the day PipeWire changes something underneath a program nobody is updating.
If you outgrow it
The graduation path is painless: Unload in NoiseTorch (restoring your plain mic), then set up EasyEffects or the filter-chain on the same machine. Nothing NoiseTorch does is persistent once unloaded, which is one of its genuinely good design decisions — it leaves your system the way it found it.