RedPanda Transcribe

Transcribe audio to text without uploading anything

Everything runs inside your browser. Your file never reaches a server — open DevTools and check the Network tab yourself.

No uploadNo sign-upNo time limitFree
Audio language: EnglishItalianGermanSpanishFrenchDutchPortugueseTurkishSwedishHebrewRussian
Audio language:English
Drop an English audio or video file here
or click to choose — MP3, M4A, WAV, MP4, MOV, WebM…
Works with audio and video. MP3, MP4, M4A and MOV have no size or length limit — they are read in slices. Other formats are handled whole, up to 1 GB or about 3 hours.

Transcript

Audio length – Processing time – Faster than real time – Data sent 0 KB
00:00
00:00
Pick a file and the transcript appears here.You can click the text to correct it.

Sample audio: made from the Google FLEURS dataset (CC BY 4.0). It is read speech, so easier to recognise than a real meeting recording.

Why "no upload" is verifiable This is not a promise, it is a property of the architecture: there is no upload endpoint. Open your browser DevTools, go to the Network tab, and transcribe a file — you will see the model being downloaded, and no request carrying your audio. The one thing stored on your device is the transcript itself, kept in this browser so an interrupted job can resume; it is deleted after 7 days and never uploaded.

How this differs from other free transcription services

What matters
Typical free service
This tool
What mattersYour file
Typical free serviceUploaded to the provider’s servers
This toolNever leaves your device
What mattersFree length
Typical free serviceUsually capped, often 10–30 minutes
This toolNo limit
What mattersFile size
Typical free serviceTypically around 1 GB
This toolNo limit for MP3 / MP4 / M4A / MOV
What mattersAccount
Typical free serviceUsually required
This toolNot required
What mattersOffline use
Typical free serviceNot possible
This toolWorks from the second visit on
What mattersAccuracy
Typical free serviceLarge cloud model — 3.8% word error rate
This toolOn-device model — 6.9% word error rate
See the measurements for every language

Accuracy depends on your material. This is enough for a usable first draft of a recording; minutes you intend to circulate still need a human pass.

Would you use a more accurate model?

We are considering a larger on-device model. It may take longer to download and run. It is not available yet.

How it works

01 Pick a file Pick a file. The browser decodes it to 16 kHz mono audio locally.
02 Processed on your device On the first visit the speech model is downloaded once (about 118 MB) and cached. Later visits load it from the cache with zero download.
03 Edit and export Recognition runs in WebAssembly on your own machine. Text appears as it is recognised.

What people use it for

Voice memosZoom recordingsVideo filesDictaphone recordingsInterviews

Questions

Is there a file size or length limit?

MP3, MP4, M4A and MOV files are read in pieces, so there is no practical limit — multi-hour recordings work. Other formats (WAV, OGG, WebM…) have to be read whole, which caps them at about 1 GB or 3 hours; convert to MP3 if you hit that.

How accurate is it?

On clean read speech in English this tool scores 6.9% word error rate, against 3.8% for Whisper large-v3 running in the cloud, measured on the same audio. Real recordings are harder than a test set: expect a usable first draft that still needs a human pass for names and technical terms.

How long does it take?

It depends on your computer. With 8 or more cores, several recognizers run in parallel at about 12–20× real time, so an hour of audio takes roughly 3–5 minutes; with 4 cores, about 7–10 minutes. Phones are much slower. Keep the tab in front: browsers can slow background tabs down considerably. Text appears as it is produced, together with an estimate of the time left.

Does it work offline?

After the first visit, yes — the model is cached in your browser.

Can I transcribe other languages?

Each page handles exactly one language, because the speech model for that language is downloaded to your browser. To transcribe another language, open that language’s page from the links at the top — no setting to change here.

What if I close the tab while it is working?

The finished part is saved on your device and offered back the next time you open the page. Pick the same file again and it carries on from where it stopped. The saved text stays in this browser only, never leaves your device, and is deleted after 7 days.