Transcribe

How to transcribe on this page

1. Read the upload note above the square. This uploads. First 3 minutes. 25 MB.
2. Drop a voice note: .m4a, .mp3, .ogg, .opus, .wav, .webm, or a short .mp4. WhatsApp voice notes (audio/ogg, audio/opus) are welcome. If Grok rejects opus, we convert to WAV in this tab first, then upload the WAV. Cap is still 3 minutes / 25 MB after convert.
3. Pick a language if you know it: Auto, Estonian, Finnish, or English. Estonian and Finnish are not on Grok's official language list. We still let you pick them. English is on the list. Auto is fine when you are unsure or the note mixes languages.
4. Click Transcribe. The file is posted to /api/transcribe on this host, forwarded to Grok (xAI), and then dropped from memory. We do not write it to disk.
5. Read the text. Copy it or download the .txt. There is no watermark and no signup wall.

WhatsApp voice notes (Estonian and Finnish)

This page exists for the voice note you cannot replay in a browser tab and do not want to type out. WhatsApp is the usual source. Estonian and Finnish landings live at /et/transcribe and /fi/transcribe — same tool, language already set, copy in that language. WhatsApp-first URLs: /et/whatsapp-haal and /fi/whatsapp-aani. Video that will not send: /compress-video/whatsapp.

On iPhone, open the chat, tap and hold the voice note, then Save or Share and get the file onto the laptop — AirDrop, Files, or Mail to yourself. iOS WhatsApp almost always gives you an .m4a (AAC). That is the easiest drop for this tool.

On Android, tap and hold the note, then Share or Save to Downloads / Files. Android WhatsApp often saves Opus in an .ogg or .opus container (audio/ogg, audio/opus). Drop that. If Grok rejects opus, this tab converts it to WAV with ffmpeg.wasm and uploads the WAV. Cap is still 3 minutes / 25 MB after convert. If the phone only offers “forward in WhatsApp”, use the three-dot menu on the note and look for Save. A forwarded note that never left the app is not a file we can read.

Estonian and Finnish voice notes are why these landings exist. They are not on Grok's official language list. We still send the file and let you pick Estonian or Finnish. The model may guess, mix in English, or wobble on street names, company names, and fillers like “noh” / “niinku”. Read the transcript before you paste it into anything official.

Group chats and speaker labels are not something we add. You get one block of text. If two people talked over each other, the model will guess. A quiet room and one speaker is the setup that works.

What actually happens to the file

The browser measures duration when it can, then POSTs the bytes to /api/transcribe. If this host has no XAI_API_KEY in the environment, the route returns 503 with { error: "no_key" } and the UI says so. If the key is set, the server holds the upload in memory, calls Grok (xAI speech-to-text), returns the text, and wipes the buffer. We do not write the voice note to disk. We do not keep a copy after the response. xAI's own retention is theirs, not ours — read their policy if that matters for the note.

That is the honest version. The file clinic never uploads. This front does. The privacy page repeats it.

Why three minutes

v1 is the first three minutes, free, no account. The client rejects a longer estimate. The server rejects a duration field over about 3.2 minutes, and anything over 25 MB. A twenty-minute meeting does not belong here. Cut it, or use a tool you run yourself. We will not unlock “the rest” behind a paywall on this page.

Some WhatsApp ogg files will not report duration in the browser. We still send them if they are under 25 MB. If Grok then refuses a long file, you will see a failure, not a locked transcript.

What this is not

This is not a meeting bot, not a live captioner, and not a store of your voice. No account, no ads on the download, no “sign in to copy”. If you need hours of audio or a folder of notes, run transcription on a machine you control. Once is for the one note you need as text in the next minute.

FAQ

Does the file leave my computer?

Yes. This server, then Grok, then we drop it. If you need local-only, use the file clinic instead.

Do I need an account?

No. The text is shown and downloadable without a login.

What if the host has no API key?

You get a clear error. We do not pretend the tool is working. Set XAI_API_KEY in the environment (see the README). Do not put a key in the repo.

Is the job counted?

A successful response increments localStorage key once:jobs on this device. That number is not sent with the audio.

Can I transcribe a whole mp4 lecture?

Not here. Extract the audio on the file clinic if you want a file, then cut the first three minutes if you still want this page.