Transcribe Audio and Video to Text, or Make SRT Subtitles
Transcribe speech in an audio or video file to text, Word, Markdown, SRT or VTT subtitles, on your device. The recording is not uploaded.
Want AI to do the finding? Try AI Audio Transcriber (sign-in required).
Drop an audio or video file to transcribeTap to choose an audio or video file to transcribe
or click to browseor tap to choose
English speech, up to 60 minutes. Not uploaded.
Large files (over 2 GB) may take longer or run out of browser memory
Transcribes speech in an audio or video file (MP3, WAV, M4A, MP4, WebM and more) into text you can read, search, correct and download as a Word document, Markdown, plain text, or SRT and WebVTT subtitles. A small English speech model runs inside your browser, so the recording is never uploaded. The first use downloads the model (about 100 MB) and keeps it in this browser so later uses start at once.
Drop an audio or video file (up to 60 minutes) and press Transcribe. When it finishes, read the transcript while the file plays, fix wrong words (or use find and replace), choose paragraphs, timestamps and a speaker guess, then copy it or download Word, Markdown, text, SRT or VTT.
English on this device. Clear speech with little background noise works best. Names, product words and numbers are the usual mistakes, so read the text once while the file plays. Timing comes from the model's own sentence timestamps and can sit a fraction of a second off. The speaker guess only looks at long pauses: it cannot tell voices apart.
- How do I transcribe an MP3, voice memo or video to text?
- Drop the audio or video file (MP3, WAV, M4A, MP4, WebM and more) on the page and press Transcribe. When it finishes you get a readable transcript in paragraphs that you can correct, search, copy, or download as Word (.docx), Markdown or plain text. The same run also gives you SRT and VTT subtitle files.
- Is my recording uploaded?
- No. A small speech model runs inside your browser, so the sound stays on your device. The only download is the model itself, about 100 MB, which your browser keeps after the first time.
- Which languages does it understand?
- On your device it understands English only; speech in another language will come out as wrong or empty text. For other languages and for long recordings, the optional higher-accuracy mode (sign in) uses our AI service, which covers 38 or more languages. It only appears on this page when it is switched on, and it sends the file for that one request without keeping it.
- Can it label who is speaking?
- Not reliably on your device. You can tick a guess that flips between Speaker 1 and Speaker 2 after a long pause, but it only looks at silence and cannot tell voices apart, so check it. Real speaker labels come with the higher-accuracy mode when it is available.
- How accurate is it, and how long does it take?
- It is a small model, so it handles clear speech well and makes mistakes on names, product words, numbers and noisy recordings. Read the text while the file plays and fix what is wrong, or use find and replace for a name that is wrong everywhere. Expect it to take about half the length of the recording, longer on a phone.
- Which files can I download?
- For the transcript: Word (.docx), Markdown (.md) and plain text, with optional timestamps at the start of each paragraph. For video editors and YouTube: SRT and VTT subtitle files. Correct any wrong words first, then download.
- How long can the file be?
- Up to 60 minutes on your device. The file is cut at the pauses and done in parts, and you can press Stop at any time and keep what is written so far. Split longer recordings and do the parts one by one. Video files work if your browser can play their sound, which is usually MP4 and WebM.
How to use Transcribe Audio to Text
Drop an audio or video file with English speech and press Transcribe; the first time your browser downloads a speech model of about 100 MB and keeps it. You can press Stop whenever you like and keep what is written. When the text appears, play the file, correct wrong words in the boxes or with find and replace, and choose paragraph breaks, timestamps and the optional speaker guess. Then copy the transcript or download Word, Markdown or plain text. The Subtitles tab gives SRT or VTT for a video editor or YouTube.
When to use it
Use it to turn interviews, meetings, lectures, podcasts, voice notes and short videos into a transcript or captions when you do not want to upload the recording to a service. It suits MP3 to text, video to text and voice to text jobs where you will read and fix the result.
When not to use it
Do not use the on-device mode for languages other than English, for songs, or for noisy recordings where several people talk at once. It is not a certified transcript for legal or medical records.
Tips for best results
Use a clear recording with little background noise. Read the text once while it plays, because the usual mistakes are names and numbers; find and replace fixes a wrong name everywhere at once. Import the SRT into your video editor and nudge the timing there if a line is slightly off.
Limitations
The on-device model is small, so names, numbers and unusual words are often wrong. Speaker labels are only a guess from long pauses. Timing comes from the model's sentence timestamps, so a line can be a fraction of a second early or late, and there is no word-by-word timing. A long recording takes about half its length to process, and the limit is 60 minutes.
Privacy
In the normal mode the recording is decoded and processed in your browser and is never sent to a server. Only the public speech model is downloaded, from FreeDailyPro, and your browser saves it so it loads at once next time. You can remove it from the page. The optional higher-accuracy mode, when it appears, sends the file for that one request; it is read in memory and not kept.
More from Audio Tools: