Real-Time Translation Overlay for Windows Audio

What it does

LiveTranslate sits between your Windows audio output and your eyes. It captures system audio via WASAPI loopback—optionally mixing in your microphone—feeds it through a voice-activity detector and an ASR engine, then streams the translation through any OpenAI-compatible LLM API into a draggable, click-through overlay. The result is a universal subtitle bar that works with any video player, browser, or voice chat without modifying the source application.

The interesting bit

The project treats the entire Windows audio mix as a broadcast source, which means it doesn’t need plugins, browser extensions, or per-app integration. It also lets you offload speech recognition to a remote GPU over HTTP, so a cheap laptop can run the overlay while a desktop somewhere else handles the heavy ASR lifting.

Key highlights

• Works with any system audio source; no player modifications or browser extensions needed.
• Supports multiple ASR backends: faster-whisper, SenseVoice, FunASR Nano, and even Anime-Whisper for Japanese game and anime dialogue.
• Compatible with any OpenAI-compatible translation API, including DeepSeek, Grok, Ollama, and vLLM.
• Optional remote ASR mode to run inference on a separate GPU machine.
• Portable build auto-downloads its own Python runtime and GPU dependencies, so non-developers can unzip and run it.

Caveats

• Windows 10/11 only; there is no Linux or macOS support.
• Translation requires an external LLM API and network access; it does not run a local translation model out of the box.
• GPU acceleration is recommended for low-latency ASR, though CPU inference is possible.

Verdict

Worth a look if you are a VTuber, streamer, or anyone who watches untranslated foreign content on Windows and wants a plug-and-play overlay. Skip it if you are on Linux, macOS, or looking for a fully offline solution that bundles its own translation engine.