Start transcribing
Sign up now to start transcribing your audio and video files with high accuracy, secure and private processing, and industry-leading features like speaker diarization and subtitle generation.
Upload a recording and get it back as text. Recordings up to 10 minutes are free without an account; with a free account you get 30 free minutes, speaker labels and faster processing, and after that you pay per minute – no subscription. Audio is half of what is uploaded here: MP3 (22 % of all files), M4A voice memos (16 %), WAV and OGG.
Any common audio format: MP3, M4A and AAC (iPhone and Android voice memos), WAV, AIFF, OGG, Opus and FLAC.
Timestamps when you need them: plain text, or SRT/VTT with timing, or JSON with every word.
Speaker labels: with a free account, interviews and meetings come back with each speaker marked.
Language detected automatically: 55 languages detected on this site so far; a recording in Spanish, French or Arabic needs no setting.
Private by default: uploads are deleted after 30 days; delete a transcript and its file earlier yourself.
Free up to 10 minutes without an account · 30 free minutes with one.
Supported formats: MP3, M4A, WAV, OGG, FLAC, AAC, AIFF · Prices
MP3 and M4A voice recordings are 38 % of uploads. Paying users’ M4A memos run 20 minutes at the median – an hour of typing, two minutes of upload.
Facts updated August 2026
On OpenAI’s published benchmark of read speech (Common Voice 15) the large-v3 model makes about one error in 11 words in English, one in 21 in Spanish, one in 18 in German and one in 17 in Portuguese and Russian. On clean studio recordings (FLEURS) it is roughly a third better. Independent testing on real meetings and calls (Open ASR Leaderboard, 2026) puts English at 7–11 % word error. transcribe.mov runs the faster “turbo” build of the same model, which is a few tenths of a point behind on English. Real files are harder than benchmarks: background noise, cross-talk and strong accents add errors, and a wrong language guess ruins a transcript.
Sources: OpenAI large-v3 release chart (Common Voice 15, FLEURS), Open ASR Leaderboard paper 2026-03-30.
MP3, M4A/AAC, OGG and Opus are compressed; WAV, AIFF and FLAC are not. The speech model works from the audio it hears, and a 64 kbps voice recording transcribes as well as a studio WAV. What matters is the recording itself: distance to the microphone, room echo and background noise, not the file type or the bitrate.
An hour of MP3 at 128 kbps is about 55 MB; the same hour as WAV is around 600 MB. Guests can upload up to 10 minutes and 50 MB, account holders up to 2 hours and 4 GB per file. Most recordings uploaded here are short – the median is 2.5 minutes and nine in ten are under 12 minutes.
Mono and stereo are both fine; nothing is lost when a stereo interview is transcribed. Recordings from a phone on the table work well in a quiet room; the further the speaker is from the microphone and the more people talk at once, the more you will correct afterwards in the editor.
Four lines from a transcribed recording as the online transcript shows them with “Show Timecodes” and “Show Speakers” switched on. The labels are generated automatically; the JSON export carries them too.
[00:00:00 --> 00:00:05] [SPEAKER_00] Thanks for making the time. Can you start with how the team is set up? [00:00:05 --> 00:00:11] [SPEAKER_01] Sure. There are six of us, and three of them joined this year. [00:00:11 --> 00:00:14] [SPEAKER_00] And what changed when they came in? [00:00:14 --> 00:00:19] [SPEAKER_01] Mostly the handover. We used to lose a day every week to it.
The TXT download contains the same text without the labels; SRT and VTT add the timing of each line.
MP3, M4A and AAC (iPhone and Android voice memos), WAV, AIFF, OGG, Opus and FLAC – the usual outputs of phones, recorders, DAWs and podcast software. If a recorder writes something else, export it as MP3 or WAV.
transcribe.mov uses Whisper large-v3 (the turbo build) – the open model that OpenAI benchmarks per language. On read speech it makes roughly one error in 11 words in English and one in 18–21 in Spanish, German, Portuguese or Russian; clean studio audio is better, noisy or overlapping speech is worse, and independent tests of real meetings land around 11 % word error. You get the whole file, verbatim, with punctuation and speaker labels (registered users); you fix names and terms in the editor.
Record close to the speaker in a quiet room – one voice at a time transcribes best. Upload the original file; don’t re-encode or compress it (phone .m4a and camera .mov are fine as they are). Clear speech matters more than file quality: accents and cross-talk cost more accuracy than bitrate ever will. Silence and music are filtered automatically.
Up to 10 minutes and 50 MB without an account, free. A free account gives you 30 free minutes and files up to 2 hours and 4 GB. Beyond that you buy minutes: 120 for $5.90, 300 for $10.90, 600 for $19.90 – one-time, no subscription, and the minutes never expire.
With an account, yes: the online transcript marks each change of speaker as [SPEAKER_00], [SPEAKER_01] and so on, and the JSON export carries the same labels. Guest transcripts and the TXT, SRT and VTT files contain the text without speaker names.
Yes. Download the transcript as SRT or VTT and every few seconds of speech carry a start and end time; the JSON export has the timing of every single word. Plain TXT has no timestamps.
The language is detected automatically – 55 languages so far on this site, with Spanish, French, Portuguese, Tagalog, German and Arabic the most common after English. A recording that mixes two languages is transcribed in the one that dominates.
The uploaded file is deleted from our servers automatically after 30 days. The transcript stays in your account until you delete it – you can delete a transcript and its file at any time. Files are processed automatically and are never shared.
For a short clip you can. Gemini takes audio and video uploads, but its free tier stops at 10 minutes of audio or 5 minutes of video per prompt and needs a Google sign-in (3 hours of audio with a paid AI Pro plan). ChatGPT has no documented audio or video transcription feature: it may try on an uploaded file, and OpenAI itself says the analysis "may not analyze the entire video or accurately interpret its audio". In both, timestamps are written by the language model, not measured, so they drift, and there are no speaker labels on uploaded files. Here the recording is transcribed in full and verbatim by a speech model, the SRT/VTT timing comes from the audio itself, speakers are labelled, files up to 10 minutes need no account at all, and your files are never used to train anything.
Facts updated August 2026
Sign up now to start transcribing your audio and video files with high accuracy, secure and private processing, and industry-leading features like speaker diarization and subtitle generation.
transcribe.mov - Free, Accurate, and Instant Transcription for Your Audios and Videos! Unlock the Power of Our AI-Powered Transcription Tool.
