José Miguel Gómez

Back to the home page

Sublingua

Live translated subtitles for any call, entirely on your laptop

This production is a private summary: description, learnings and stack are shown, the code stays private.

Draft. Written from the program notes and pending the author’s review.

Status
Coming soon
Type
Application
Running since
January 2025
Role
Architect and sole engineer
Domains
speech recognition, translation
Cast
Python, Faster Whisper, Litellm

Context and problem

Desktop interpreter and live subtitles for Windows.

Assumption: professionals in multilingual teams follow calls in a language they read better than they hear, and the available live-subtitle tools send the audio of the call to a third party.

Sublingua shows live translated subtitles for any call and runs the speech recognition entirely on the laptop; a cloud model is optional and used only for summaries.

Screens

How it was built

1 audit

Source: Trysac/11b-sublingua-v2-interprete UI/UX audit (no development report), dated 21 November 2026.

What the project taught

  • Local by design is a product constraint, not an optimisation: choosing a model size that runs in real time on a CPU decides what the product can promise.
  • Quantised small models are enough for live subtitles when latency matters more than the last point of accuracy.
  • Keeping the cloud optional and limited to summaries draws a line users can understand: the call stays on the laptop, the recap may leave it.

Case study in preparation

The full case study adds these sections when it is published:

  • Constraints
  • Architecture diagrams (C4 context and containers)
  • Three key decisions
  • What went wrong
  • Results with dated evidence
  • Demo video
  • What I would do differently

In the project mesh

  • Forgeline, low effort: shared resources llm-gateway; a standard interface exists
  • Resolvo, low effort: shared resources llm-gateway; a standard interface exists

Highlights

local first, speech to text, privacy