Live recording tool
Real-time speech to text, free in your browser
Press record, talk, press stop. The transcriber below returns punctuated, capitalized text a few seconds later — no account, no install, three free runs a day. The engine is NVIDIA's Nemotron 3.5 ASR streaming model, served by this independent site. One honest note before you start: the transcript appears after you stop recording, not as live captions while you speak.
Activation is for existing license holders. New purchases are paused.
Your transcript will appear here with punctuation and capitalization.
How to turn speech into text in real time
The whole flow takes under a minute the first time and about ten seconds every time after that.
- Click record in the transcriber above and allow microphone access when your browser asks. The permission applies to this site only.
- Pick a language or leave detection on auto. The model identifies the spoken language on its own, and it keeps up even if you switch languages mid-recording.
- Choose a speed mode. Balanced is the sensible default; the table below explains when to pick the other three.
- Speak normally. A free recording can run up to 5 minutes.
- Press stop. The punctuated, capitalized transcript appears within seconds.
- Copy the text, or download it as a .txt file.
What real-time means on this page
Worth being precise about, because tools that rank for real time speech to text often mean quite different things by the phrase. The model underneath is a genuinely streaming architecture: a 0.6B cache-aware FastConformer-RNNT network that processes audio in chunks as short as 0.08 seconds. But this site does not draw captions on your screen while you talk. You record, you stop, and the full transcript arrives seconds later.
That trade makes it right for some jobs and wrong for others. Right: dictating a first draft, capturing meeting minutes, turning a lecture or interview into searchable text, voice journaling. Wrong: caption overlays for a live stream, broadcast subtitling, or accessibility captions during an event, where words must appear as they are spoken. If you need that second category, this is not your tool — better to say so here than waste one of your three free runs finding out.
Four speed modes for real-time speech to text
The mode selector changes the chunk size the streaming model uses. Smaller chunks mean less audio context per decision, so you trade a little accuracy for a shorter wait after you press stop.
| Mode | Model chunk size | Pick it when |
|---|---|---|
| Best accuracy | 1.12 s | The recording matters: interviews, lectures, anything you will publish or quote. |
| Balanced | 0.56 s | Everyday dictation and meeting notes. The practical default. |
| Faster | 0.32 s | Quick voice notes where speed beats the last percent of accuracy. |
| Lowest latency | 0.08 s | You want the shortest possible gap between stopping and reading. |
Microphone habits that improve every transcript
Clean input helps any ASR model more than any setting does. Five habits worth keeping:
- Stay close to the microphone — within arm's length. Distance adds room echo, and echo is one of the most common causes of dropped words.
- Avoid clipping. If your recording app shows a level meter hitting red, back off; distorted peaks are unrecoverable for any model.
- Kill background music. ASR models separate speech from steady noise reasonably well, but competing vocals and lyrics confuse them.
- Prefer a headset or lapel mic over a laptop's built-in array when you can. Consistent distance beats raw mic quality.
- In a group setting, place the mic near whoever talks most. This tool does not label speakers, so position is your only lever.
What people use real-time speech to text for
- First drafts by voice — most people speak around three times faster than they type, and the punctuation arrives already in place.
- Meeting minutes: record the discussion, stop, and paste the text into your notes before everyone has left the room.
- Lectures and talks, up to 5 minutes per run — record longer sessions in segments.
- Interviews, with the honest caveat that output is one continuous text without speaker labels, so mark turns yourself.
- Voice journaling: talk through the day, stop, and keep the .txt. Nothing to set up and no account tying entries to you.
40 languages, detected automatically
The transcriber accepts 40 language locales, including English (US and UK), Spanish, French, German, Portuguese, Japanese, Korean, Russian, Hindi, Arabic, and Mandarin Chinese (zh-CN). You can set the language explicitly, or leave it on auto and let the model decide from the audio itself.
Auto-detection also survives code-switching. If you start a note in English and drift into Spanish halfway through, the model follows and keeps transcribing — useful for bilingual households, language classes, and international calls recorded on speakerphone.
Free limits, stated plainly
Free means 3 transcriptions per day per connection, up to 5 minutes and 20 MB each, with no account and no email. You can also upload files (MP3, WAV, M4A, AAC, OGG, OPUS) instead of recording live. New Pro purchases are paused; existing Creem license keys still work.
What this tool does not do, so you are not surprised later: no speaker diarization, no word or segment timestamps, no SRT or VTT export, no video files, no batch uploads, and no public API. Output is plain punctuated text you copy or download — nothing more, and nothing hidden behind a paywall reveal.
Frequently asked questions
Is this really real-time speech to text?
The model is a true streaming architecture that processes audio in chunks as small as 0.08 seconds, but this site delivers the transcript after you press stop — typically within a few seconds — rather than painting captions while you speak. For dictation and notes that distinction rarely matters; for live captioning it does, and this tool is not built for live captioning.
Can it caption a live stream, webinar, or Zoom call?
No. There is no live caption overlay and no way to pipe audio in continuously. If you can record the audio and stop, you can transcribe it here; if you need text on screen while people are still talking, use a dedicated live-captioning service instead.
Does microphone recording work on phone browsers?
Yes. Modern mobile browsers — Safari on iOS and Chrome on Android — support in-browser microphone recording. Allow mic access when prompted and the flow is identical to desktop, with the same free limits of 3 runs a day and 5 minutes per recording.
Which languages does it support?
40 language locales with automatic detection, including English, Spanish, French, German, Portuguese, Japanese, Korean, Hindi, Arabic, and Mandarin Chinese. Auto-detect handles a recording that switches languages partway through.
Is it free, and are existing Pro licenses supported?
Free gives you 3 transcriptions a day, up to 5 minutes and 20 MB each, with no account required. Existing Pro licenses retain their original allowance and can activate in up to 3 browsers. New purchases are paused.
Does the transcript include timestamps or speaker names?
No. Output is one continuous block of punctuated, capitalized text. There are no word timestamps, no SRT or VTT subtitle export, and no speaker labels, so multi-person recordings come out as a single stream of text.
Can I upload a recording instead of using the mic?
Yes. The same transcriber accepts MP3, WAV, M4A, AAC, OGG, and OPUS uploads under the same limits — 5 minutes and 20 MB per file on the free tier.