Video transcript
Drop in a recording and watch the words come in. Your computer does the listening, so the video stays with you.
Or drop it here. MP4, MOV, WebM, MKV or an audio file.
On our 30-minute test video, the sentences came through clean enough to search and quote. The names didn't, so fix those before you share it. Got a YouTube link instead of a file? This tool needs the video file itself.
Is the transcript good enough?
We tested it on a 30-minute video of an office-hours call: a computer voice reading a short script 22 times, with names, places and products in it. That's one clear voice and no background noise.
The transcript had 5,685 words; the script had 5,698. Numbers come out as digits: “two hundred” becomes “200”. A few sentences were missing or worded differently in up to 3 of the 22 repeats.
Names are where it slips. Monika came out “Monica” all 44 times. Skool came out “school” and Circle came out “circle” every time. Kajabi was spelled right in 15 of 22. Before you send a transcript to anyone, search it for each person and product in the video and fix the spelling.
0:01I'm Marcel, and Monica said: Monika is here with me today.
7:25He moved his course off Kajabi last spring and wants to know what to do with 200 hours of Zoom recordings.
7:51Someone asked whether to use school said: Skool or circle said: Circle for the community,
How long will it take?
On our test laptop, an Apple M5 Pro, the whole 30-minute video took 82 seconds, from choosing the file to the finished transcript. That's with WebGPU, which lets the browser use the computer's graphics chip. In a browser without it, the tool runs on the computer's main processor instead: a 5-minute cut of the same video took 65 seconds, so an hour would take about 13 minutes.
The first time, your browser downloads the speech model, about 209 MB with WebGPU. The browser keeps it, so your next video starts sooner. Keep the tab open while it works.
Which file to download
Pick by where the transcript is going.
- A doc, an email or your notes: plain text. One paragraph you can paste anywhere.
- Quoting a moment, or finding it again later: with timestamps. Each line starts with the minute and second it was said.
- Captions for a video you upload somewhere: SRT, a subtitle file where every line has a start and an end time.
- Captions on your own website's video: VTT, the caption format web pages use for video.
[0:00] Welcome back to Office Hours. [0:01] I'm Marcel, and Monica is here with me today.
Welcome back to Office Hours. I'm Marcel, and Monica is here with me today.
1 00:00:00,000 --> 00:00:01,840 Welcome back to Office Hours. 2 00:00:01,840 --> 00:00:04,880 I'm Marcel, and Monica is here with me today.
WEBVTT 00:00:00.000 --> 00:00:01.840 Welcome back to Office Hours. 00:00:01.840 --> 00:00:04.880 I'm Marcel, and Monica is here with me today.
One video gets a transcript. A library needs answers.
A transcript makes one video searchable for you. It doesn't help a member with a question at 11 at night. They'd have to know which of your videos to open. Most programs live with that: weekly office hours, a library nobody searches, and AI nobody trusts.
Bold hosts your recordings and answers your members from them, with the minute you said it. When the recordings don't cover a question, it says so. At HRT University, members ask 20 times as many questions as before: 25 a week in live Q&A, 500 a week on Bold. See how it works, or how it fits a coaching program. If you only need the sound from a video, use the free audio extractor.
Questions
Can AI transcribe a video for free?
Yes. This page runs Whisper, OpenAI's open speech model, in your browser. It costs nothing, needs no account, and the version it runs is free to use under the Apache 2.0 license.
Does it say who's speaking?
No. It writes down what was said, line by line, without names. If you need speakers, add them yourself after you download it.
How long a video can I use?
We don't set a limit, and we've tested 30 minutes. The tool holds the video's sound in your browser's memory while it works, so a very long video on a phone with little memory can run out.
Does it work in other languages?
Whisper was trained on speech in many languages, not only English. We've only tested it in English.