How to use Audio & video transcription
- 1
Add a recording
Drop an MP3, M4A, WAV, MP4 or MOV file, or switch to Record audio and talk into your microphone.
- 2
Start transcribing
Base suits most recordings; Small copes better with accents and noise. Leave the language on auto-detect or set it.
- 3
Read, play and search
Follow the highlighted words during playback, click a word to jump there, and search for names or topics.
- 4
Save it
Copy the text or download TXT, DOCX, SRT, VTT or JSON with word timings.
Limits and Pro
Up to 10 minutes of audio per file or recording (longer files are transcribed for their first 10 minutes), files up to 500 MB. The speech model downloads once: 43 MB (Tiny), 80 MB (Base) or 252 MB (Small).
Up to 600 minutes per file and files up to 4 GB. See Pro
About Audio & video transcription
Transcribe is a speech-to-text tool for interviews, lectures, meetings, podcasts, sermons and voice notes. You choose a recording or capture one with your microphone, and the open-source Whisper model writes down what was said, with punctuation and capital letters, in any of 99 languages. Long recordings are processed in 30-second windows, so the text appears while the rest is still being worked on, and a cut-off word at a window edge is re-read from the next window instead of being lost.
The result is a reading view rather than a wall of text: sentences are grouped into paragraphs wherever the speaker pauses, each paragraph can show its start time, and every word carries its own timestamp. Press play and the word being spoken is highlighted; click any word to hear it in context. Search finds a phrase across the whole transcript and steps through each match.
Downloads cover the usual next steps: plain text (with or without timestamps) for notes, a Word document for editing and sharing, SRT or VTT if the recording is going to be published as a video, and JSON with per-word start and end times for developers. The transcript does not label who is speaking, and it is not a live captioning tool — recording comes first, transcription second.
Privacy
Recordings stay on your computer or phone: files are opened locally, microphone audio is recorded into memory on this page, and Whisper runs in a background worker without sending audio anywhere. Close the tab and the recording and transcript are gone. Only the speech model is fetched from Hugging Face once and kept in your browser's cache.
Features
Nothing is uploaded
Decoding and recognition both happen in this browser tab, so confidential calls and interviews stay with you.
99 languages
Automatic language detection, plus translation from any supported language into English.
Word-level timestamps
Every word is timed, so you can click to play and follow along with live highlighting.
Record from your mic
Capture a voice note, interview or dictation and transcribe it right away.
TXT, Word, SRT, VTT, JSON
Export the format you need — documents, subtitles or structured data for your own tools.
Search the transcript
Find every mention of a name or phrase and jump between matches.