# Egrekci

> Website: https://egrekci.lucapixel.com/ (Türkçe) · https://egrekci.lucapixel.com/en/ (English)

> Egrekci is a Windows app that removes background noise from your microphone in real time with AI (AI noise cancellation, a Krisp alternative). It works as a virtual microphone in Discord, Zoom, Microsoft Teams, Google Meet, OBS and every other app. It runs fully offline on the CPU: no GPU, no internet, and the user's voice never leaves the PC.
> Türkçe: Egrekci, mikrofondaki arka plan gürültüsünü yapay zekâ ile gerçek zamanlı silen bir Windows uygulamasıdır (gürültü engelleme / noise cancellation, Krisp alternatifi).

## Key facts

- Platform: Windows 10 and Windows 11, 64-bit
- Current version: 26.1.2 (year-based versioning: 26 = 2026)
- Download: [EgrekciSetup-26.1.2.exe](https://egrekci.lucapixel.com/downloads/EgrekciSetup-26.1.2.exe), 23 MB, one installer that includes the virtual microphone driver
- Removes: keyboard clicks, desk taps, mouse clicks, fans, air conditioning, traffic and room hum
- Voice effects (applied after noise cancellation): Studio (default, warm and full), Podcast (bass-heavy, tight dynamics), Clear (bright, forward highs), Natural (no effect)
- Controls: system tray, global hotkey Ctrl+Alt+N, starts with Windows, voice test with record/playback, warning when the microphone is muted or off
- Languages: Turkish and English; the app follows the Windows display language and can be switched in Settings
- Privacy: all processing is local; no account, no network access
- Setup: install, restart Windows once, then Egrekci makes itself the default microphone; if an app needs it, choose "Egrekci Mikrofon (VB-Audio Virtual Cable)" as the input device

## Technology

- Virtual microphone: VB-CABLE by VB-Audio Software, donationware, installed by the Egrekci setup ([product](https://vb-audio.com/Cable/), [donate](https://shop.vb-audio.com/en/win-apps/11-vb-cable.html)). Egrekci is not affiliated with VB-Audio.
- Noise model: DPDFNet8 48 kHz full-band speech enhancement (Ceva, Apache-2.0), a DeepFilterNet descendant with dual-path RNN blocks
- Inference: ONNX Runtime on the CPU, about 5 ms per 10 ms of audio on a mid-range desktop CPU (Ryzen 5 5600)
- App: written in Rust; WASAPI audio; high-pass filter and automatic gain before the model; parametric EQ and compressor after it

## Pages

- [Türkçe ana sayfa](https://egrekci.lucapixel.com/): features, voice effects, setup, compatibility, FAQ
- [English page](https://egrekci.lucapixel.com/en/): the same content in English
- [Third-party licenses](https://egrekci.lucapixel.com/THIRD-PARTY.txt)