Actually live: your speech pipes to our own AI while you talk
While you record, the page slices the microphone stream into short segments — about seven seconds each — and pipes every segment straight to our GPU servers, where the same Whisper-class model that powers the whole site transcribes it immediately. The text lands in the live feed a few seconds behind your voice and keeps flowing for as long as you keep talking. No browser speech service, no third-party dictation API — the live text is real Whisper output from our own hardware.
Because it only needs microphone recording, live mode works in every modern browser — Chrome, Edge, Safari, and Firefox, desktop and phone. Each audio segment is deleted from our servers the moment its text comes back.
The live feed vs the final transcript
Segment-by-segment transcription has one built-in limit: the model sees seven seconds at a time, so a word that straddles two segments can come out clipped, and punctuation across segment boundaries is guesswork. That's why the flow ends with one more step: the full recording is kept in your browser while you speak, and pressing Transcribe runs it through the model as a single take — full context, proper punctuation, timestamps, the editor, and TXT/SRT/VTT exports.
Use the live feed to confirm it's keeping up and to grab text the moment it's spoken (there's a copy button on the feed); use the final pass for anything you'll keep. For audio you already have as a file, upload mode on any tool page is the shorter path.
Frequently asked questions
Is this actually live, or recorded and sent afterwards?
Actually live: recording and transcription overlap. The page ships each ~7-second slice of audio to our GPUs while you're still speaking the next one, so text accumulates during the recording, not after it. Expect the feed to run a handful of seconds behind your voice — the time it takes a slice to finish, upload, and transcribe.
Why is the live text a few seconds behind me?
Three small delays stack up: a slice has to finish being recorded (up to 7 seconds), travel to the server, and pass through the model (about a second on our GPUs). That puts the feed roughly 8-10 seconds behind your voice. Dictation-style speech with natural pauses feels smooth; the delay is only noticeable when you watch for it.
Which browsers does live mode work in?
All modern ones — Chrome, Edge, Safari, and Firefox, on desktop and phones. It relies on standard microphone recording rather than a browser speech service, which is also why Firefox works here even though it lacks built-in dictation.
Where does my voice go while live transcription runs?
To our servers only, and honestly described: live mode streams your audio to us in short segments as you speak — that's what makes it live — and each segment is deleted immediately after its text comes back. Nothing goes to Google, Apple, or any third-party speech service. The full recording stays in your browser until you press Transcribe, and anonymous transcripts auto-delete within 24 hours.
Can I use it to follow a meeting or lecture in the room?
As a personal aid, yes — put the device near the speaker and the feed tracks clear speech well, about ten seconds behind. Know the limits: one microphone, so distant voices and crosstalk degrade it, and it isn't a certified captioning service for accessibility compliance. Press Transcribe at the end and the full-context pass usually recovers what the live feed fumbled.
Which languages work live, and should I pick one?
The same 90+ languages as every transcription on this site — it's the same model. One tip specific to live mode: pick your language in the selector instead of leaving auto-detect on. Detection runs per segment, and seven seconds isn't much evidence, so pinning the language keeps the feed from wavering on short or accented phrases.