Live Spanish ⇄ English Translator

Listens to spoken Spanish, transcribes it on your device with a Whisper speech model, and shows a running English translation beside it. The audio is transcribed entirely in your browser — it never leaves the device. Only the recognized text is sent to your chosen LLM for translation, using your own API key.

Direction

Translation provider

Translation model

API key — stored only in this browser

Transcription engine

Speech model — runs on-device, downloaded once

Use WebGPU

Cloud transcription model

OpenAI API key — for cloud transcription

Vocabulary hints (optional) — proper names, places, and archaic or regional words the speaker uses, so the transcriber expects them

Home GPU server URL — your streaming server, exposed with tailscale serve

Audio streams to your own machine over your tailnet; transcription (Whisper large-v3) and translation (local LLM) run there. No API keys needed. The device must be connected to Tailscale.

Export filename root — the downloaded file is named root-YYYYMMDD-HHMM.md

Mic sensitivity — move right if it misses quiet speech, left if background noise triggers it

🔒 Your API key is saved only in this browser (its localStorage) and sent directly from your device to the LLM provider — never to this website's server. Speech recognition runs fully on-device, so your audio is never uploaded anywhere. Don't enter your key on a shared computer.

Idle mic level

Spanish

English

Diagnostics

If something fails, tap Copy diagnostics and paste the result to Claude. The log stays on your device; anything that looks like an API key is redacted automatically.

Live log endpoint (optional) — if set, the log is POSTed here so it can be read remotely