Your turn.
Their language.
Choose who is speaking. Record their words, then stop to see and hear the translation.
Wear headphones or a headset to reduce echo. This experimental interpreter translates after short speech pauses. It is instructed to translate into the other language without repeating the original. Accuracy and cost may differ from After speaking. Avoid speaking over each other. Translated audio is AI-generated.
Voice gender is an experimental preference. French speech uses a France French accent.
Recorded translation · Hover over a translated sentence for 🔊 replay. First generation is billed; repeats use cached audio. Original audio/text available below each turn. Audio is AI-generated.
Audio stays on this device until you click Stop and translate. The microphone is off during translated playback.
Text and audio play as they arrive after Stop.
Set app allowance
This starts a new allowance from now. Saved on this app’s server; shared across your browser tabs.
GPT-Live voice only, at US$0.05/min. App estimate, not your OpenAI credit balance. TTS replays and other modes are excluded. Failed connection attempts may not be captured. Display only; this does not enforce a spending cap.
Conversation
Available after a translation finishes. Replay sends nothing to OpenAI.
Test with a recording
Loading a file stays local. The buttons below send it to OpenAI in the selected direction.
GPT-Live diagnostic log
Separates recognized microphone text (INPUT) from generated speech text (OUTPUT), before local formatting. Stored only in this tab; no extra API calls or audio recording. Reset on the next GPT-Live session, Clear, or page refresh. Older entries are discarded if the log fills.
No GPT-Live transcript events yet.
How this version works
GPT-Live mode uses direct WebRTC audio tracks between your browser and OpenAI. Our server exchanges the connection offer and answer, keeps the API key private, and cleans up abandoned sessions. Wait for Connected and listening before speaking. The recording test streams the selected file instead of using your microphone.
Synchronous mode now uses one general Realtime interpreter with instructions to translate between the selected languages and never repeat the original language. It responds after short speech pauses and streams its translated voice and transcript. Stop finishes pending input/output; End now cancels immediately. Sessions stop after 10 minutes. Gender settings apply only to After speaking.
Tap Speak to record the tourist in the selected tourist language, or Listen to record the local in the selected local language. Tap the same button again to stop recording and send the audio for translation into the other language.
Only one side can record at a time. Both controls wait while translation and playback finish, so the microphone cannot pick up translated audio. There is no automatic mode or language-detection download in this version.
A recording can be up to 30 seconds. At the limit the microphone stops and the recording stays local until you click Stop and translate. Discard recording deletes the unsent recording. Translations use OpenAI; the server keeps your API key private.
After Stop, we recognize the original speech, translate its text, and generate speech from that translation. Expand Original transcript to check the recognized words. No separate spacing-check request is used.
Last microphone settings
No microphone has been opened.