Feature request: translate non-English speech to English in the transcript

[matthiaslesage]

Context

Thanks for quill! Having everything stay on-device makes it very valuable.

I'm not a native English speaker. My meetings happen in my own language, but
the output I want out of them, transcripts I feed to an LLM to code are much more useful to me in English. Today that means a second manual translation step after transcript.md is written.

Proposal

Support a "translate to English" mode alongside plain transcription, e.g.:

{
  "transcription": {
    "enabled": true,
    "engine": "whisper",
    "task": "translate",
    "source_language": "auto"
  }
}

Happy to help!

[upsoftt]

If you'd rather have the feature than build it: Dictator does this already - free recognition in ~100 languages locally on the GPU, AI translation into 198 and 49 localy (any direction, original kept next to the translation, not whisper's English-only task: translate), plus speaker labels by voice rather than by track. Windows/macOS/Linux and phones: dictator.my

[matthiaslesage]

Thanks @upsoftt, I looked into it.

It's a capable product, but it is not what I am looking for: its wide-coverage translation runs through cloud models with character quotas, and only a subset of languages are handled offline. quill's appeal is that the whole pipeline stays on the machine, and I'd rather not give that up for the translation step. Closed-source subscription tooling is also a different bet than a single Swift binary I can read myself.

@digimata, no pressure on timing, but is this something you'd consider for the roadmap, or is it explicitly out of scope? Asking mainly because the README already mentions a WhisperKit engine as a planned fallback, so I wanted to know whether translation could ride along with that work or whether you'd rather keep the core narrow and leave it to on_stop.

Either answer is useful. Thanks again for building this. Very few tools do fully local transcription this well!