Voice translator
Say it in your language, and your phone says it in theirs. Take turns for a two-way conversation at a taxi rank, a hotel desk or a market stall.
Getting clear results
- Speak one or two complete sentences per turn, not a paragraph.
- Hold the phone about a hand's length from your mouth, away from wind.
- Names and addresses: say them slowly, then check the written transcript.
- Listening
- Writing down words
- Translating meaning
- Speaking it back
Machine translation. Good for getting the gist of signs, menus and everyday documents. Not suitable for legal, medical, immigration or other official use; use a certified human translator for those. Recordings (max 30 s) are sent once for transcription and translation, never stored.
Three steps between your voice and theirs
1. Speech to text. The recording is transcribed by a dedicated transcription model (Google's Gemini transcribe via Vercel AI Gateway). You see exactly what it heard, which is the first thing to check if a translation looks wrong.
2. Text to meaning. The transcript is translated as spoken language, so "could you do me a favour" becomes a natural request rather than a word-for-word calque.
3. Meaning to voice. A text-to-speech model reads the translation aloud. The bar under the buttons lights up each stage as it happens.
Questions
How does the voice translator work?
Press the button for the language being spoken and talk for up to 30 seconds. The recording is transcribed by a speech-recognition model, translated, and read aloud in the other language. Both the transcript and the translation appear on screen so you can check them.
Can two people have a conversation with it?
Yes. Set "You speak" and "They speak", then take turns: you press the pink button, they press the dark one. Each turn is translated in the right direction and played back automatically. Hand the phone over or hold it between you.
Is it real-time?
It is turn-based rather than simultaneous: you finish a sentence, then hear the translation a few seconds later. That is how human interpreters work in conversation too, and it keeps each translation complete.
Which languages can I speak?
Speech recognition covers dozens of widely spoken languages, and translation targets more than 150. Accuracy is best for common languages and clear speech; heavy background noise, fast overlapping speech and rare dialects reduce it.
Are my recordings saved?
No. Each clip is sent once for transcription and translation and is not stored. The conversation on screen disappears when you close the page.
Can I use it at a doctor or pharmacy?
It can help with simple, practical exchanges, but do not rely on it for diagnoses, dosages or consent. Ask for a professional interpreter; many hospitals provide one free of charge.