Real-time voice processing for Windows that does not touch your GPU, phone home, or leave a service running in the background. It cleans up your microphone, shapes your voice, and hands the result to Discord, OBS, or a game as if it were a second mic.
The node editor with the Parallel Vibe preset loaded: the cleaned voice splits into dry, compressed, and saturated paths, then recombines through a Mix node.
Voicemod is a lot of software for what most people actually use it for: a clean vocal chain and noise reduction. That convenience costs VRAM, a constant connection to someone else's API, and a steady drip of upsells. Formant is the part worth keeping. It runs a small RNNoise model and a handful of well-tuned effects on the CPU, with nothing reaching out to the network, and it routes into anything through a virtual audio cable.
It began as a straight replacement for that daily workflow and then grew a node editor, so the simple case stays simple and the deep case is there when you want it.
your mic -> Formant's chain -> your headphones (so you can hear yourself)
\-> a virtual cable -> Discord / OBS / game
One mono input, processed once, sent to two places at the same time: your own ears, and a virtual microphone that every other app reads from. Point an app's microphone at the cable and you are finished.
The chain. High-pass, RNNoise denoise, gate, de-esser, compressor, 3-band EQ, saturator, limiter, and gain. Enough to take a raw mic to a finished voice, with bypass on every node and a global bypass for quick before-and-after. The Mixer tab lays the chain out as a channel strip with live meters.
A real node editor. Pan, zoom, and wire it however you like. A Mix node sums its inputs and any output can fan out, so you can build parallel compression, blend-in saturation, and dry/wet balance, not just a single line of effects.
Your plugins, hosted. Formant finds the VST3 plugins already installed on your machine, lets you drop them in as nodes, opens their own editor windows, and saves their settings inside your presets.
Mic control that stays out of the way. Voice-activity gating by default, plus push-to-talk, toggle, and always-open, all on global hotkeys you can rebind or clear.
It lives in the tray. Close the window and the audio keeps running. One instance only, an optional start-with-Windows toggle, and it reopens on the exact chain you left behind.
No glitches. A drift-compensated resampler on the output path keeps capture and playback clocks in step, so long sessions do not slowly drift into crackle.
- Windows 10 or 11.
- A virtual audio device so other apps can read Formant as a microphone. VB-CABLE (free) or Voicemeeter both work.
- Headphones or speakers, if you want to monitor yourself.
Grab the MSI from the Releases page for a normal install, or the portable zip if you would rather not run an installer.
To build from source instead (per-user, no admin needed):
pwsh packaging/install.ps1 # build release, install, add a Start Menu shortcut
pwsh packaging/install.ps1 -Desktop # also add a Desktop shortcut
pwsh packaging/uninstall.ps1 # remove (add -Purge to also delete config and presets)
pwsh packaging/build-portable.ps1 # build a portable zip into dist/
pwsh packaging/build-msi.ps1 # build the MSI into dist/The script installer puts Formant in %LOCALAPPDATA%\Programs\Formant; the MSI
installs to Program Files. Either way, config and presets live in
%APPDATA%\Formant.
- Open the Setup tab. Pick your microphone as the capture device, your headphones as the monitor output, and your virtual cable as the second output.
- In Discord, OBS, or your game, set the microphone to that same virtual cable.
- Shape your sound in the Mixer and Nodes tabs, or just load a preset.
Two presets ship with the app to get you going:
- Standard Vocal is the classic single chain: high-pass, denoise, gate, de-esser, compressor, EQ, makeup.
- Parallel Vibe splits the cleaned voice three ways into a dry path, a heavily compressed parallel path, and a saturated path, then blends them back together through a Mix node. It is a good tour of the parallel routing.
cargo test # core DSP, engine, and preset tests
cargo run -p formant -- --seconds 5 # headless: stream mic through the chain and report stats
cargo run -p formant # the full GUIThe whole DSP and graph core is testable without a microphone, which is how the math stays honest.
crates/core(formant-core): DSP, the graph engine, routing, presets, and session state. Pure and fully unit-tested without an audio device.crates/audio(formant-audio): the Windows WASAPI backend, device enumeration, and global hotkeys.crates/vst3(formant-vst3): VST3 discovery and hosting.crates/app(formant): the egui interface, the tray, and the glue between them.
See SPEC.md for the design and architecture in more detail.
Formant is free software under the GNU General Public License, version 3 or later. See LICENSE for the full text.
Copyright (C) 2026 Ethan Belanger.
It is GPLv3 because it hosts VST3 plugins through Steinberg's interface definitions, which are themselves available under GPLv3 (or a separate Steinberg commercial license).
Formant builds on open-source work by others, including RNNoise for noise reduction and egui for the interface. See THIRD-PARTY-NOTICES.md for credits. VST is a trademark of Steinberg Media Technologies GmbH; Formant is an independent project and is not affiliated with Steinberg.