Kommz Gamer

I've been building Kommz Gamer for the past year, but the idea is actually 10+ years old. Back then I was a pure geek trying to learn programming as a self-taught dev, but I eventually gave up on it.

Last year I was playing Valorant and met an English speaker and a Spanish player in a match. Impossible to communicate. That's when this project came back to me. I told myself: "Now there's AI, let's try it. I have nothing to lose." Long story short, I got addicted. 13-16 hours a day for 4 months, and I had a working version online.

What it does:

🌍 Real-time voice translation across 20+ languages (Deepgram Nova-2)

Transcribes your voice, translates it, and speaks it back - all in a few seconds.

🎮 Auto game detection in 2-3s (audio fingerprint) with automatic profile switching

Identifies the game and applies the right audio profile automatically.

🎧 Voice Focus V3: voice isolation, noise reduction, auto calibration

Filters out keyboard noise, background music, and other players' voices.

📺 OBS overlay for streamers

Displays translations as subtitles on your stream.

🧬 Optional voice cloning (your own voice, translated into any language)

Record 30 seconds of your voice - your teammates hear YOU speaking their language.

🐟 Fish Audio Premium → a second TTS engine option, bring your own API key

More expressive voices with emotional markers.

⚡ <300ms latency in Turbo Mode (Edge TTS only → voice cloning trades speed for quality)

Fast mode for competitive play, quality mode for voice cloning (~1.5s GPU synthesis).

🛡️ Anti-cheat safe → no DLL injection, no memory reading, works like OBS/Discord

Compatible with Vanguard, Ricochet, and BattleEye.

📱 Mobile Command Center to control everything from your phone

Change language, adjust mic sensitivity, all from your phone.

What I learned building this:

🐍 Python monolith → modular Flask architecture

Refactored from a ~11K line monolith into 13 modular Flask blueprints (~81 routes). That was the biggest technical challenge.

🔨 Nuitka onefile build

Compiling Python to native C. qt.conf, PySide6 DLLs, memory management - the Windows build pipeline was one of the hardest parts.

🎧 Audio on Windows is a nightmare

WASAPI loopback, CABLE Input, Voicemeeter, ASIO - each setup is different. That consumed most of the year.

🧠 Voice cloning pipeline

XTTS v2 on Modal GPU cloud, direct /v1/synthesis endpoint. Benchmarked live: ~1.2-1.6s GPU synthesis, ~4-6s round-trip total, WER 0-7% in French.

📊 Live benchmarks

- STT (Deepgram Nova-2): ~370ms

- Translation (DeepL): ~320ms

- TTS (XTTS Modal GPU): ~1.2-1.6s

- WER: 0-7% (tested in French, clear speech)

⚙️ Tech stack

Python + Flask (13 blueprints, ~81 routes), Nuitka onefile build, HTML/JS frontend in PyWebView, Deepgram STT, DeepL translation, Edge TTS + XTTS voice cloning on Modal GPU, WASAPI loopback + CABLE Input for Windows audio capture.

🎥 Demo video (French audio with English subtitles): https://www.youtube.com/watch?v=6BZbSO--iCY

The project is open-source, split into two repos:

🖥️ Kommz Gamer (the desktop app): https://github.com/KommzAI/Kommz-Gamer

🐟 Kommz Voice (the voice cloning engine): https://github.com/KommzAI/Kommz-Voice

⚡ Free 24h trial (no card, full access), then plans starting at €10/month: https://kommz.fr

💬 Discord: https://discord.gg/XpVCYktJ4B

🚀 Also on Product Hunt: https://www.producthunt.com/products/kommz-gamer-v5-3

Solo dev from France, built everything alone. AGPLv3 - contributions welcome. AMA about the monolith refactoring, Nuitka build, or the Windows audio pipeline.