VoxGo
Real-time voice translation overlay for PC gamers.
Capture Windows system audio, transcribe speech locally with Whisper, and show translated subtitles in a lightweight game overlay.
Built for real PC game voice, not meeting rooms.
VoxGo focuses on overseas servers, Discord voice, squad comms, and lightweight subtitle mirroring while you keep playing.
System audio capture
Captures the sound Windows is already playing through WASAPI loopback. It is not microphone capture by default.
Local Whisper ASR
Speech recognition runs locally with faster-whisper. The Lite package downloads the model on first use; the Full package includes multilingual small/base models.
Flexible translation
Use OpenAI-compatible providers, local compatible services, or Google Cloud Translation Basic v2.
Game overlay
A transparent, always-on-top overlay shows translated subtitles without injecting into the game process.
Mobile mirror
Open the built-in QR code to mirror subtitles on a phone or a second device on your local network.
Diagnostics included
Audio testing, translation testing, latency metrics, and feedback templates make troubleshooting less mysterious.
Download VoxGo
Start with Lite if your network can download the Whisper model. Use Full if you want multilingual models bundled, or Full-CUDA / GPU if you are on NVIDIA and want GPU recognition immediately.
Lite package
Smaller download. On first run, VoxGo downloads the speech recognition model and guides you through API and audio tests.
Download VoxGo v0.4.0 LiteFull package
Includes multilingual Whisper small/base models and no CUDA DLLs. English-only models download on demand after enabling Pure English.
Download VoxGo v0.4.0 FullFull-CUDA / GPU package
Includes multilingual Whisper small/base models plus CUDA DLLs. English-only models still download on demand after enabling Pure English.
Download VoxGo v0.4.0 GPUHow to use
Check translation and audio before joining a match. Most subtitle issues come from the wrong Windows output device.
Download
Download Lite, Full, or Full-CUDA and unzip the archive.
Run VoxGo
Launch VoxGo.exe and complete the first-run setup.
Test API
Choose OpenAI-compatible or Google Translation and run the API test.
Test audio
Select a system audio / loopback device and make sure the level meter moves.
Play
Move and lock the overlay, then use hotkeys or the tray icon in-game.
FAQ
Short answers for the things people usually ask before trusting an overlay tool.
Does VoxGo record my microphone?
Not by default. VoxGo is designed to capture Windows system audio output, such as game voice, Discord, or video audio.
Does it inject into games?
No. VoxGo is a standalone desktop app. It captures audio through Windows devices and displays subtitles in a transparent always-on-top window.
Why do I hear sound but see no subtitles?
The selected audio device may not match your current Windows output. Use the audio test meter in setup or settings, then choose the loopback device that moves when game or Discord sound plays.
Which languages are supported?
The current public UI focuses on English and Chinese: English to Chinese and Chinese to English.
Lite, Full, or Full-CUDA?
Lite has no model and no CUDA DLLs. Full includes multilingual Whisper small/base but no CUDA DLLs. Full-CUDA includes those multilingual models and CUDA DLLs for NVIDIA users. English-only models download on demand only after enabling Pure English.
Privacy / Safety
VoxGo keeps the risky parts explicit: no game process injection, system audio capture instead of microphone by default, local Whisper ASR, and user-controlled translation providers.
No game injection
The overlay is an independent window. It does not patch, hook, or modify game clients.
Local transcription
Whisper transcription runs on your PC. Translation requests go to the provider you configure.
Open source
The code is available on GitHub, including packaging scripts, update manifest, and runtime behavior.
Try it, break it, report real game cases.
VoxGo needs feedback from different games, headsets, languages, accents, and Windows audio setups. The built-in feedback template includes version, device, provider, latency, and log paths.