Proposal: continuous sentence-by-sentence translated captions for an audience

[yharby]

First, thank you for FluidVoice. I am using the prebuilt macOS app and it works very well.

Use case

I sometimes speak English to an audience that does not understand English. I would like to use FluidVoice as a lightweight live captioning/interpreter tool, with a configured OpenAI or OpenAI-compatible model correcting the transcript and translating it into the audience language.

The current Toggle activation mode and Live Preview help, but the preview shows the raw speech transcript, while AI enhancement is applied only after I stop dictation. Pressing Right Control or Option after every sentence interrupts a presentation.

Proposal

Would the maintainers and community be interested in an optional Continuous Caption / Interpreter Mode?

A possible flow could be:

1. The user explicitly starts one captioning session.
2. FluidVoice keeps listening and optionally shows the raw partial transcript.
3. End-of-utterance detection or a configurable silence pause finalizes a sentence or short spoken segment.
4. The finalized segment is sent through the selected AI Enhancement prompt/model for correction and translation.
5. The translated result is appended to a dedicated, resizable, audience-friendly caption window or overlay.
6. Recording resumes automatically for the next segment until the user pauses or ends the session.

Possible settings might include:

• Source and target language
• End-of-utterance or silence threshold
• Raw partial text on/off
• Dedicated large-text window, overlay, or both
• Number of recent caption lines to keep visible
• A small rolling context window so names, pronouns, and terminology translate consistently
• Local-only transcription with a clear warning when finalized text is sent to a cloud AI provider

Relationship to the existing live-dictation idea

This seems related to Discussion #427, but the goal is slightly different. Instead of inserting and correcting every partial word inside an arbitrary focused text field, this mode could append finalized translated segments to a FluidVoice-owned caption view. That may avoid some of the text-replacement and cancellation difficulties discussed there.

Privacy and safety

I think this should always require an explicit start, show a persistent recording indicator, provide an immediate pause/stop control, and clearly explain whether transcript text is being sent to a cloud provider. It should never begin listening automatically when the app launches.

I am not asking for a roadmap commitment; I mainly wanted to see whether other users would find this useful and whether the architecture could support it. I would be glad to test such a workflow.

Would a dedicated continuous caption/interpreter mode be useful to others? Would sentence-level or short-utterance segmentation be the better starting point?

[altic-dev]

So there is a live meeting transcription that might come out soon, and that could possibly be a starting point to do what you want Fluid Voice to do.

Once we have that set up, then we can be like: can you translate what is being transcribed live right now (live captioning)