MicForge

A real-time microphone processor for Windows. It captures your mic, runs a deep, fully configurable DSP chain, and outputs to a virtual audio device so any app (Discord, OBS, games, browser, calls) uses the cleaned-up, shaped signal as its "microphone".

Written in C# / .NET 10 / WPF using NAudio for WASAPI capture and render. All DSP is hand-written (no VST/plugin host), and the dark UI is a hand-rolled theme (no control-library dependency).

See ARCHITECTURE.md for the code layout and design, and ROADMAP.md for where it's going.

Highlights

• 23-stage DSP chain, every stage toggle-able, tunable live, and drag-to-reorder.
• Crafting tab — stack "voice character" cards (Deep, Bright, Radio, Megaphone, Robot…) with intensity sliders to design a voice with no technical knobs; JSON-configurable, with a built-in preview voice so you can hear it without talking.
• Meters/analyzer tab — live spectrogram, signal-flow diagram, and a full loudness suite (momentary / short-term / integrated LUFS, LRA, true-peak).
• Presets — built-in voice profiles + your own preset folder that auto-loads; plus quick EQ-curve presets on the equalizer.
• Reliability — device auto-reconnect, follow-the-default-mic, an audio watchdog, crash logging, and single-instance launch.
• Daily use — global hotkeys, push-to-talk / push-to-mute, a mute overlay, a live tray icon, and minimize/close to tray with an optional "Start with Windows".

Signal chain

Input gain ─▶ Auto gain ─▶ High-pass ─▶ Hum remover ─▶ De-plosive
  ─▶ Noise suppression (RNNoise) ─▶ Gate ─▶ Expander ─▶ De-click
  ─▶ Keystroke suppressor ─▶ De-reverb ─▶ Echo remover ─▶ Equalizer
  ─▶ Compressor ─▶ Multiband ─▶ De-esser ─▶ Saturation ─▶ Exciter
  ─▶ Voice changer ─▶ Comfort noise ─▶ Limiter ─▶ Output gain ─▶ Loudness ─▶ virtual mic

The order above is the default — you can rearrange stages by dragging their cards. Everything runs internally at mono, 32-bit float, 48 kHz.

The tabs

• Processor — the chain as cards (Sliders ⇄ Graphs view). Interactive EQ curve + spectrum, compressor transfer curve, gate/de-esser/limiter meters, in/out level meters, clip indicators, mic-health readout.
• Crafting — character cards, filterable by category (Tone / Polish / Fun / FX), with a live preview voice and per-card info + "technical peek".
• Meters — spectrogram, signal-flow, loudness suite.
• Shortcuts — configure global hotkeys and push-to-talk.
• Settings — startup/tray, self-monitoring device, reliability & diagnostics, profile export/import, presets folder, undo/redo.

Requirements

1. Windows 10/11 (x64).
2. VB-CABLE (free) — creates the virtual mic. After installing, MicForge sends audio to "CABLE Input", and in your app you pick "CABLE Output" as the microphone.

The installer is self-contained (bundles the .NET runtime), so no separate .NET install is needed to run it. To build, you need the .NET 10 SDK.

Usage

1. Install VB-CABLE and reboot.
2. Run MicForge.
3. Input = your real mic. Output = CABLE Input (VB-Audio Virtual Cable) (auto-selected if found).
4. Click Start, speak, and tune. Pick a preset to get going fast.
5. In Discord/OBS/etc., set the microphone to CABLE Output.

To monitor yourself while tuning, enable Monitoring in Settings and choose your headphones (expect a little latency — normal for the two-device WASAPI path).

RNNoise AI noise suppression

A 64-bit rnnoise.dll — built from the official xiph/rnnoise source — is bundled in native/ and copied next to the exe on build, so the Noise Suppression stage works out of the box (BSD license in native/RNNOISE-LICENSE.txt). You can also point the app at a different rnnoise.dll via the card's Load… button; the loader validates the exports before enabling the stage.

Where your data lives

Everything is under %AppData%\MicForge\ so it survives uninstall/reinstall:

• micforge.json — your full setup (chain params, devices, stage order, hotkeys, crafting, last preset)
• craftcards.json — Crafting card definitions (editable; built-ins refresh on update)
• presets\*.json — your saved presets (auto-loaded into the dropdown)
• samples\*.wav — your own Crafting preview voices
• logs\micforge.log — rolling log + crash capture

Building & packaging

# build / run
dotnet build -c Release
dotnet run   -c Release

# self-contained publish (what the installer bundles)
dotnet publish -c Release -r win-x64 --self-contained true

# installer (Inno Setup 6): installer/MicForge.iss  ->  MicForge-Setup-x.y.z.exe

The bundled preview voice clip is a CC BY-NC excerpt — see assets/VOICE-SAMPLE-LICENSE.txt for attribution.

Notes

• Capture and render run on two independent device clocks; a 500 ms input buffer plus overflow discard absorbs the drift. Great for voice; not sample-accurate.
• The Echo Remover targets slap-back / speaker-bleed echo of your own voice (single-channel, no far-end reference) — it is not full duplex acoustic echo cancellation.

License

PolyForm Noncommercial 1.0.0 — free for noncommercial use.