中文

LiveTranslate: real-time audio translation for Windows

Captures system audio and optional microphone input, runs ASR, translates via an LLM API, and shows results in a transparent overlay.

Verified

Core facts

Read by a script from the official API, not rewritten by a model.

License
MIT
Languages
Python
Maintainer
TheDeathDragon
Verified

What it is

LiveTranslate is real-time audio translation for Windows. It captures system audio (WASAPI loopback) and optional microphone input, runs ASR, translates via an LLM API, and displays results in a transparent overlay. The real-time pipeline runs system audio, VAD, ASR, LLM translation and overlay. The transparent overlay is always on top, click-through and draggable, with 14 color themes.

Main features

  1. It supports several ASR engines: faster-whisper, SenseVoice, FunASR Nano and Anime-Whisper.

  2. Remote ASR offloads speech recognition to a GPU machine over HTTP.

  3. It works with OpenAI-compatible APIs such as DeepSeek, Grok, Qwen, GPT, Ollama, vLLM, etc.

  4. The streaming translation display outputs translation character by character in real time, and per-model settings include streaming, structured output (JSON), context history and disable thinking.

  5. Microphone mix-in optionally mixes microphone input with system audio for ASR, and the VAD uses 32ms chunks plus Silero VAD with adaptive silence detection.

  6. CUDA acceleration provides GPU-accelerated ASR inference, auto model management offers a setup wizard and ModelScope / HuggingFace dual sources, and the built-in benchmark compares translation model speed and quality.

Getting started

  1. On its first run, the portable build auto-downloads a portable Python 3.12 and installs GPU-aware dependencies, with no Python installation needed.

  2. Double-clicking the install batch file runs the installer, which detects Python 3.10–3.12 and auto-installs it via winget if missing.

  3. The installer creates a virtual environment, auto-detects an NVIDIA GPU and lets the user choose CUDA or CPU PyTorch.

  4. Double-clicking the start batch file launches the program, and double-clicking the update batch file pulls the latest code and updates dependencies, auto-installing Git via winget if missing.

  5. On first launch, a setup wizard appears to choose the download source (ModelScope / HuggingFace) and cache path.

  6. The main UI appears when ready.

From the README · bash
git clone https://github.com/TheDeathDragon/LiveTranslate.git
cd LiveTranslate
From the README · bash
python -m venv .venv
.venv\Scripts\activate

# PyTorch (choose one)
pip install torch torchaudio --index-url https://download.pytorch.org/whl/cu126  # CUDA
pip install torch torchaudio --index-url https://download.pytorch.org/whl/cu128  # CUDA (RTX 50xx)
pip install torch torchaudio --index-url https://download.pytorch.org/whl/cpu    # CPU only

# Dependencies
pip install -r requirements.txt

# Launch
.venv\Scripts\python.exe main.py

Who it is for

The required OS is Windows 10/11. The Python requirement is 3.10–3.12, or the portable build can be used. A recommended GPU is NVIDIA with CUDA 12.6, and Blackwell GPUs like RTX 50xx require CUDA 12.8.

The network requirement is access to a translation API. It works with system audio from videos, livestreams and voice chat. No player modifications are needed.

On first launch, the Silero VAD and SenseVoice models download automatically (about 1GB). The project is licensed under the MIT License.

Sources

This page is written from the official sources below. Open them to check the original text.

Verified ·

This page is written by AI from the official sources above and checked automatically. This site has not tested it. Found a mistake? Tell us.