Video to text: get the transcript of any video

How to transcribe video to text

1. Drop an MP4, MOV or WebM file on the box above. Set the spoken language if you know it; Auto-detect works for most clips.
2. Wait while your browser pulls out the audio (a progress bar shows this step) and the speech model writes it down.
3. Skim the lines on the right and fix anything wrong. Click a line to hear that moment again.
4. The Export tab is already open. Press TXT to download the transcript.

There's no copy button. Open the downloaded .txt in any text editor, select all, and paste it into your doc. The steps are the same for MP4 to text as for a MOV straight off your phone.

What the video transcript looks like

The TXT file is the caption lines, one per line, in order. No timestamps, no numbering, no speaker names. A lecture clip comes out like this:

Today we're looking at how glaciers move.
Most of the movement happens at the base,
where meltwater lets the ice slide.
The top layer is brittle, which is why
you get crevasses near the surface.

Each line is a caption, so lines stop at about 42 characters and break at pauses, not at the end of a sentence. Vertical videos get shorter lines of about 22 characters. Punctuation and capitals come from the speech model, so sentences are already marked.

For prose, join the lines. In Google Docs or Word, use find and replace to swap line breaks for spaces, then press Enter wherever a new paragraph should start. If you need to know when each line was spoken, download the SRT too; it has the same text with start and end times. Our SRT generator page explains that format for uploads.

Uses for a video transcript

• Show notes and descriptions for a podcast or YouTube episode, with the main points pulled from what you actually said.
• A first draft of a blog post from a talk or tutorial. Editing a transcript down is quicker than writing from a blank page.
• Quotes. Search the text for the line you remember, then click it in the editor to check the exact wording against the audio.
• Finding a moment in a long recording, such as a meeting or a webinar, with Ctrl+F (Cmd+F on a Mac) instead of scrubbing.
• Study notes from a recorded lecture.

If you want the words on screen, not in a document, you need captions rather than a transcript. Use the subtitle generator for timed subtitle files, or auto captions to burn styled captions into the video itself.

Getting a more accurate transcript

The model is Whisper large-v3 turbo, which handles everyday speech well. Most errors come from the recording, not the model. A few things help:

• Choose the language from the menu for clips under a minute, or when the video opens with music. Auto-detect can guess wrong on both.
• Lower or remove background music before exporting the video, if you edited it yourself.
• A mic close to the speaker beats a camera mic across the room, every time.
• Check names, product terms and numbers. These are where speech recognition slips most, and a wrong figure in a quote is worse than a typo.

Every fix you make in the editor ends up in the TXT.

Languages

Whisper recognizes more than 50 languages and detects the language on its own. The menu lists 22 to pick explicitly, including English, Spanish, Hindi, Portuguese, French, German, Arabic, Bengali, Japanese, Korean and Chinese. The transcript comes out in the language that's spoken. There's no translation, so a Spanish video gives you Spanish text.

Long recordings

Video transcription is free here, with two limits: each file can be up to 30 minutes, and each visitor gets 20 minutes of transcription a day (1 credit per started minute). So in practice one transcription can cover up to 20 minutes. For anything longer, split the recording into parts of about 19 minutes. Any video editor can trim a file into pieces. If you have FFmpeg, this cuts a file into 19-minute parts without re-encoding:

ffmpeg -i lecture.mp4 -c copy -f segment -segment_time 1140 -reset_timestamps 1 part%02d.mp4

The cuts land on the nearest keyframe, so a part can run a few seconds long, which still fits. Transcribe the parts in order and paste the TXT files together. A 90-minute recording comes out as 5 parts, and at 20 minutes a day that's about 5 days of transcription. For long recordings every week, a paid transcription service will suit you better.