Activate LiveAudio in four real steps.

ACTIVATION SEQUENCE

1. 01

   // get

   Get LiveAudio — installer or source.

   Most people just download and run. Two real paths: the graphical installer (recommended), or running from source with uv if you are a developer.

   Linux: needs libportaudio2

   // installer

   Installer (recommended)

   Windows: download LiveAudio-Setup-1.2.0.exe and run it. Linux: download the tarball, extract it, and run ./liveaudio-launcher. No Python to install — the app provisions its own Python 3.11.

   linux x86_64 Copy

   # Linux: extract and launch
   tar -xzf LiveAudio-1.2.0-linux-x64.tar.gz
   ./liveaudio-launcher
   

   Windows: not code-signed

   // source

   Run from source (uv)

   Clone the repo and let uv create the environment. Choose exactly one torch extra: --extra cpu for CPU, or --extra cu121 for NVIDIA CUDA. Then run liveaudio.

   uv · developers Copy

   git clone https://github.com/plynte-labs/LiveAudio.git
   cd LiveAudio
   # pick exactly one torch extra:
   uv sync --extra cpu
   uv sync --extra cu121
   uv run liveaudio
   

   Linux needs libportaudio2 (sudo apt install libportaudio2).

2. 02

   // first-run

   First run provisions everything.

   The first launch downloads Python, dependencies and models, then caches them. Plan for roughly 400 MB on CPU or 2.5 GB on CUDA. Models download on first transcription. Every later start is instant and fully offline.

   First run: internet required

   Footprint: ~400 MB (CPU) / ~2.5 GB (CUDA). After that, fully offline. Portable mode: drop an empty portable.marker next to the launcher to keep everything in a local data/ folder.

3. 03

   // detect

   LiveAudio auto-detects your GPU.

   A GPU is optional — the CPU works. On launch, LiveAudio checks for a usable NVIDIA CUDA GPU and falls back to CPU automatically. You can always override the choice.

   • CUDA NVIDIA driver ≥ 525 and VRAM ≥ 4 GiB → CUDA is used.
   • CPU No usable GPU → CPU is used automatically. No setup needed.
   • override You can force CPU or CUDA yourself before starting.

   GPU optional; CPU works. CUDA needs driver ≥ 525, VRAM ≥ 4 GiB.

4. 04

   // go

   Pick a profile, then START SYSTEM.

   Choose an FPS-aware profile, pick your audio device and CPU/CUDA + model, click Apply changes, then START SYSTEM. Speak — captions appear and stream to OBS over the local WebSocket.

   • Fast low latency
   • Balanced recommended
   • Quality higher accuracy, more VRAM
   • Stable Streaming reduces GPU load while gaming

   Editing a built-in profile creates a Custom one. Apply changes to activate.