Speak any language.
Type English.
Tellaflow's Live Translation lets you think and speak in whatever language your brain is operating in — and outputs clean English text into any app, entirely offline.
See it in action
How it works
Enable Live Translation
Open Settings → Transcription and toggle Live Translation on. Make sure you have a Whisper model downloaded — Parakeet does not support translation.
Press your hotkey and speak
Speak in your native language. Tellaflow captures your audio and passes it to Whisper running locally on your Mac.
Whisper translates on-device
Whisper's multilingual model was trained specifically for translate mode — it produces English output from any supported input language in a single inference pass.
English text appears immediately
The translated English text is injected into whatever app has focus. No copy-paste, no round-trip to a server, no translation API key needed.
Why Whisper can do this without an API
OpenAI trained Whisper on 680,000 hours of multilingual audio, with translation supervision included in the training objective. The model was explicitly built to produce English output from any supported language — not as an add-on, but as a core capability. When Tellaflow runs Whisper in translate mode, it calls a flag that was always there.
Why offline translation matters
Every other tool routes translation through a cloud API. Here's why that's a problem.
No translation API
Services like DeepL and Google Translate charge per character. Whisper does it locally for free — it's built into the model.
Works on a plane
Even if you need to write in English mid-flight from a multilingual meeting. No WiFi, no problem.
No extra latency
Translation happens in the same inference step as transcription. There's no second pass, no second request.
Who uses Live Translation
A product manager in Bangalore thinks in Hindi, documents in English. She dictates meeting notes in Hindi and gets English output instantly. Her whole team can read them.
A linguist reads academic papers in French and German. He annotates them verbally in those languages; Tellaflow produces English notes he can share with English-speaking collaborators.
Many multilingual people switch languages mid-sentence naturally. Live Translation handles this — Whisper's model was trained on mixed-language audio.
Someone learning Japanese speaks sentences aloud and immediately sees the English translation. A fast feedback loop for checking comprehension without leaving the app they're studying in.
65 input languages
All supported by Whisper's multilingual model. Output is always English.
Accuracy varies by language, model size, and audio quality. Whisper large and medium perform best for less common languages and accented speech.
Your language.
Their English.
Nobody else's business.
Free, offline, runs on your Mac. No translation API charges. No one listening.
Get started → Model setup guide