Skip to content
OpenCaptions

Self-hosted Whisper captions with Docker

OpenCaptions runs speech recognition on your own machine and turns the result into burned-in animated captions or subtitle files. Nothing leaves the server unless you choose a provider that needs it.

Terminal window
curl -O https://raw.githubusercontent.com/leogaudin/opencaptions/main/docker-compose.yml
docker compose up -d

Images are published for amd64 and arm64, so it runs on a Linux server, a Mac with Apple silicon or a Raspberry Pi.

Which model

Section titled “Which model”
  • Parakeet Ultra, the default, covers 25 European languages and is fast on a CPU.
  • Whisper covers about 100 languages, and on a GPU it is the fast one.

The speech models page lists them, and the benchmarks show the accuracy and speed we measured, with the method.

What you get

Section titled “What you get”

Word-level timing, a timeline to correct it, nineteen looks, and exports to MP4, WebM, ProRes, SRT, VTT and JSON.

Questions

Section titled “Questions”

Does it need a GPU? No. A GPU speeds up transcription, and it is optional.

Is there an API? Yes. Scripts can use it with an API key, and another OpenCaptions server can transcribe for this one.