How to Transcribe a Video to Text
Four ways to turn a video into text, what is actually happening when you do it, and how to check the result without reading it twice.
The Scriptorium team · · Updated 12 September 2026
The short version: upload the video to a transcription tool, wait, then correct the result while listening to the audio. That last step is the one people skip and the one that decides whether the transcript is any good.
The longer version depends on what you have and what the transcript is for. Below are the four routes that actually exist, what is actually happening when a video becomes text, and how to handle the situation you are actually in — your own file, someone else's YouTube video, a recorded meeting, or a page you cannot download from.
What is actually happening
Converting a video to text means reading the speech on its sound track and writing it out as words. The picture is never involved. A video file is a container holding at least two streams — a picture track and a sound track — and a transcription tool opens the container, ignores the picture entirely, decodes the audio, and runs speech recognition over it.
Two consequences follow, and both are useful to know before you upload anything:
- Resolution does not matter. A 4K file and a 480p export of the same talk produce the same transcript. If you are choosing which copy to upload, pick the one with the better audio, not the better picture.
- Anything without speech produces nothing. Music, room tone, sound effects and silence come back empty. That is correct, not a failure.
Route 1: upload the file to a transcription tool
This is the default and it is the right default. You give a tool the video file, it reads the audio track and gives you back text, usually with a timestamp on every line.
- Open a transcription tool — video to text here, if you want to try it now.
- Drop the video file in. MP4, MOV, MKV, WebM, AVI, M4V, WMV, FLV, MPG, MPEG, TS and 3GP all work; there is no need to convert the container or extract the audio first. Converting an MKV to an MP4 before uploading achieves nothing except a second generation of compression on the audio, which can only make the transcript worse.
- Set the language if you know it. Auto-detection is good but can be thrown by a recording that opens with music or noise.
- Wait. Transcription runs faster than real time, so a recording usually takes a fraction of its own length, though you may queue behind other work.
- Read the transcript with the audio playing and fix what is wrong.
- Export it in the format the next step needs.
Free tiers exist and are genuinely usable for short files. Here it is one file up to 15 minutes and 25 MB with no account at all.
Route 2: use captions the platform already made
If the video is already on YouTube, a transcript may exist for free. YouTube auto-captions most public videos, and on desktop you can open that text directly: click …more under the title, scroll to the bottom of the description, and click Show transcript.
It is rougher than a dedicated pass and has no speaker labels, but it costs nothing and takes ten seconds. The YouTube transcript page covers this in detail, including what to do when the panel is missing.
Route 3: type it yourself
Worth naming because it is sometimes right. For a two-minute clip where every word has to be exact, typing it while you listen is faster than uploading, waiting and correcting. For anything longer, it is not: manual transcription runs at several times the length of the recording for most people.
Route 4: pay a human service
Human transcription services charge per audio minute and turn work around in hours or days. For legal proceedings, medical records, research that will be published, or heavily accented and overlapping speech, this is still the accurate option and the money is well spent. For everything else it is overkill.
Which situation are you actually in?
The four routes above cover the general case. Here is the fastest path for the specific situations that come up most:
The video is a file on your computer or phone
Route 1, directly — drop it into a transcription tool. If the file is too large, extracting the audio usually brings it under a size limit without changing the transcript at all.
It is your own video, already on YouTube
Download it from YouTube Studio (Content, then the three-dot menu next to the video, then Download), then use Route 1. The extra step is worth it when the transcript matters: you get speaker labels, line-by-line editing and proper subtitle files, none of which YouTube's own captions give you.
It is someone else's video on YouTube
Route 2, above — YouTube's own transcript panel. That is the only route to someone else's video without downloading it yourself.
It is a recorded meeting
Check the meeting tool first. Zoom, Teams and Google Meet all offer recording transcripts, sometimes only on paid plans, and if one already exists you have saved yourself a step. If not, download the recording and use Route 1 — meeting recordings are a good case for speaker labels, since the value is usually in who said what rather than just what was said.
It is embedded on a website you do not control
Check whether the page already has a transcript published below the player — many do, particularly for talks, lectures and podcasts. If not, and you cannot download the video, there is no clean route, and that is the honest answer.
You only need part of it
Trim first. If you want the ten minutes where a specific topic came up, cutting to that section before uploading means a shorter wait, a smaller file, and a transcript you do not have to search through.
How to check a transcript quickly
Do not read it top to bottom. Play the audio and read along. Errors announce themselves because the words on screen stop matching the words in your ears, and in a good tool clicking a line jumps the audio to that second.
Errors cluster. Check these first and you will catch most of them:
- Proper nouns — names of people, companies, products and places.
- Numbers, dates, prices and units.
- The opening seconds, before the system has any context.
- Any passage where two people speak at once.
- Domain vocabulary: medical, legal, technical or internal shorthand.
Getting a better transcript before you record
Almost all of the quality is decided by the audio, and almost none of it by the video. A 4K file and a 480p file of the same talk transcribe identically. What changes the result:
- Microphone distance. Closer is better, and it beats every other factor.
- Room echo. Soft furnishings help; bare rooms smear words together.
- One voice at a time. Overlapping speech is the hardest case there is.
- Background noise. Close the window, turn off the fan.
How accurate is it?
Nobody should quote you a single accuracy figure, including us. Automatic transcription of one person speaking clearly into a close microphone is very good — good enough that correcting it is much faster than typing it. The same system on a recording with two people talking over each other, a noisy room, a strong accent or a lot of specialist vocabulary is meaningfully worse.
The failure mode worth knowing about is that it does not degrade into obvious nonsense. It produces fluent, confident, plausible sentences that are wrong, which is harder to catch than gibberish would be. That is why you check a transcript against the audio rather than reading it on its own.
Common problems
The transcript is empty
Either there is no speech on the track, or the video has no audio stream at all. Play the file with your volume up and check.
It came back in the wrong language
Auto-detection made its decision from the opening seconds. If the video starts with music, an intro sting or silence, set the language explicitly and run it again.
The file is too large
Extract the audio and upload that instead. An audio-only version of a video is a small fraction of the size and produces exactly the same transcript.
Names are spelled wrong
Expected. A speech model has no way to know how a name is spelled unless the context makes it obvious. Fix the first occurrence, then use find and replace.
Which format to export
This is the decision people get wrong, usually by exporting plain text and then discovering they needed timestamps.
| You want to… | Export |
|---|---|
| Paste it into a document | Plain text (.txt) |
| Cite the moment something was said | Timestamped text |
| Add subtitles to a video | SRT or WebVTT |
| Publish show notes or a report | Markdown |
| Process it in code | JSON with word timings |
| Work with it in a spreadsheet | CSV |
If you are unsure, export timestamped text. You can always strip timestamps out later; you cannot add them back in.