Formant processes your voice in real time on Windows. It uses the CPU only. It sends nothing to the network and it runs no background service. Formant cleans your microphone signal, shapes your voice, and sends the result to Discord, OBS, or a game as a second microphone.
The node editor with the Parallel Vibe preset. The clean voice splits into a dry path, a compressed path, and a saturated path. A Mix node combines the three paths again.
Most people use Voicemod for two functions. They want a clean voice chain and noise reduction. Voicemod charges VRAM for those two functions. It also keeps a connection to a remote API, and it asks you to buy paid features.
Formant keeps only the part that matters. It runs a small RNNoise model and a set of tuned effects on the CPU. No part of Formant connects to the network. A virtual audio cable carries the output to any other application.
Formant started as a replacement for this daily workflow. It then got a node editor. The simple case stays simple. The complex case is available when you want it.
your microphone -> the Formant chain -> your headphones (you hear yourself)
\-> a virtual audio cable -> Discord / OBS / game
Formant takes one mono input and processes it one time. It sends the result to two places at the same time. The first place is your headphones. The second place is a virtual microphone that other applications read. Set the microphone of an application to that cable and the setup is complete.
The chain. The chain contains a high-pass filter, RNNoise denoise, a gate, a de-esser, a compressor, a parametric 3-band EQ, a saturator, a limiter, and gain. This is enough to make a finished voice from a raw microphone signal. Every node has a bypass switch, and a global bypass gives you a quick comparison. The Mixer tab shows the chain as a channel strip with live meters.
The processing aims for the quality of commercial plugins, not only for correct math. The compressor follows loudness instead of the waveform. It uses a soft knee and a program-dependent release. The limiter looks ahead, so it reduces the gain before a peak instead of clipping the peak flat. The limiter also catches peaks between samples. The saturator runs oversampled, so its harmonics cannot alias into content unrelated to your voice.
The de-esser attenuates a band around your sibilance. It does not attenuate everything above a crossover frequency. A harsh "sss" comes under control and your voice keeps its air.
The node editor. Pan, zoom, and connect the nodes as you want. A Mix node sums its inputs, and any output can feed more than one node. You can build parallel compression, blended saturation, and a dry/wet balance. The chain is not limited to a single line of effects.
VST3 plugin hosting. Formant finds the VST3 plugins on your machine. Add them as nodes and open their own editor windows. Formant saves their settings inside your presets.
Microphone control. Voice-activity gating is the default. Push-to-talk, toggle, and always-open are also available. All four modes use global hotkeys that you can change or clear.
Tray operation. Close the window and the audio continues. Formant permits one instance only. An optional setting starts Formant with Windows. Formant reopens with the chain that you left.
No dropouts. A drift-compensated resampler on the output path holds the capture clock and the playback clock in step. Long sessions do not drift into crackle.
- Windows 10 or 11.
- A virtual audio cable, so other applications can read Formant as a microphone. VB-CABLE (free) and Voicemeeter both work.
- Headphones or speakers, if you want to monitor your own voice.
Download the MSI from the Releases page for a standard install. Download the portable zip if you do not want to run an installer.
To build from source, use these scripts. They install per user and need no administrator rights.
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 Formant (add -Purge to delete config and presets)
pwsh packaging/build-portable.ps1 # build a portable zip in dist/
pwsh packaging/build-msi.ps1 # build the MSI in dist/The script installer puts Formant in %LOCALAPPDATA%\Programs\Formant. The MSI
installs Formant in Program Files. Both keep config and presets in
%APPDATA%\Formant.
- Open the Setup tab.
- Set the capture device to your microphone.
- Set the monitor output to your headphones.
- Set the second output to your virtual audio cable.
- In Discord, OBS, or your game, set the microphone to that same cable.
- Shape your sound in the Mixer tab and the Nodes tab, or load a preset.
Formant includes five presets:
- Standard Vocal is the basic single chain. It has a high-pass filter, denoise, a gate, a de-esser, a compressor, an EQ, and makeup gain.
- Broadcast is tighter and more present. It sounds like a news reader.
- Warm is fuller and softer. It adds parallel compression for body and gentle saturation in place of hard control.
- Podcast stays even and clear over long sessions. It has tighter noise control and enough makeup gain for streaming loudness targets.
- Parallel Vibe splits the clean voice into three paths: a dry path, a heavily compressed path, and a saturated path. A Mix node combines the three paths again. This preset shows you how parallel routing works.
All five presets are restrained. A preset that sounds impressive at the first word usually becomes tiring after a minute. You also cannot remove heavy processing later in the chain.
cargo test # core DSP, engine, and preset tests
cargo run -p formant -- --seconds 5 # headless: stream the microphone through the chain and report stats
cargo run -p formant # the full GUIYou can test the complete DSP and graph core without a microphone. This keeps the math honest.
crates/core(formant-core): DSP, the graph engine, routing, presets, and session state. This crate is pure and its unit tests need no 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 icon, and the code that connects them.
For more detail about the design and the architecture, see SPEC.md.
Formant is free software under the GNU General Public License, version 3 or later. For the full text, see LICENSE.
Copyright (C) 2026 Ethan Belanger.
Formant uses GPLv3 because it hosts VST3 plugins through the Steinberg interface definitions. Steinberg supplies those definitions under GPLv3 or under a separate commercial license.
Formant uses open-source work from other people. This includes RNNoise for noise reduction and egui for the interface. For credits, see THIRD-PARTY-NOTICES.md. VST is a trademark of Steinberg Media Technologies GmbH. Formant is an independent project and has no affiliation with Steinberg.