Answers about local, free, real-time captions.

Yes. All audio processing and transcription run on your own machine with no telemetry, and internet is needed only on first run to download transcription models — after that it works fully offline. v1.5.1 is a native Rust binary with no Python runtime at all.

v1.5.1 focuses on Windows 10/11 x64; Linux users stay on the legacy Python v1.2.7 tarball with microphone capture, which needs the libportaudio2 package (sudo apt install libportaudio2). System-audio loopback (WASAPI) is Windows-only.

Yes. LiveAudio is free and open-source under the MIT license, with no subscription and no API key — you only pay your own electricity, and hardware is not included.

No. The CPU works on its own; NVIDIA CUDA is optional but recommended and is auto-detected, needing driver 525 or newer and at least 4 GiB of VRAM.

No API key, ever — and no account either. Internet is required only on first run to fetch transcription models; once those are cached, LiveAudio runs entirely offline.

Low and tunable — well under a second on a typical setup — using FPS-aware presets (Fast / Balanced / Quality / Stable Streaming). It is real-time and sub-second, not the few-milliseconds figure marketing pages sometimes claim.

Cloud ASR APIs send your audio off your machine and bill per minute, while browser caption tools depend on a remote service; LiveAudio does everything locally with no cloud and no per-minute cost. It broadcasts clean subtitle JSON over a local WebSocket (ws://127.0.0.1:8765) that OBS — or any localhost client — can consume.