Transcripts, and why they are not media

A transcript is a document about a video

A video file is millions of encoded frames and a stream of sound samples. A transcript is a sequence of sentences somebody typed or a machine produced from listening. One is media and the other is a document, and the difference decides everything on this page.

The size difference makes the point on its own. An hour of speech at a watchable quality runs to hundreds of megabytes as video. The same hour written out is a few tens of kilobytes, small enough to paste into an email.

A transcript is timed or plain text of spoken words, not the video file itself. This page explains what Analyze can return for a watch address and when you should take an audio row instead if you plan to produce text yourself.

Auto-generated captions differ from human-reviewed transcripts. Expect errors on names and technical terms either way, and quote carefully.

Where the words on a YouTube transcript come from

YouTube builds its transcript panel from the caption track attached to that video. Where a channel supplied written captions, the panel shows what a person typed. Where YouTube produced captions by listening to the audio, the panel shows machine output with the errors that come with it.

That is why one transcript reads like a manuscript and another reads like a raw dictation. The quality was decided when the caption track was made, long before anybody opened the panel.

What the results list can answer for a watch address

Paste a YouTube watch address here, press Analyze, and you receive rows describing the published picture and sound files. Each row shows a size in megabytes and, for picture, a height figure. Every one of those rows is media.

No row holds the talking as text, because YouTube does not publish the transcript panel as a media file. A page reading the video files finds video files. If reading the words is the whole job, this list is not the place the words live.

Taking the audio when you plan to write it out yourself

There is one route where the results list matters to a transcription job. Speech recognition software on your own device needs a recording to work from, and an audio-only row gives you that recording in the smallest form available.

Copy the watch address, press Analyze, and take the row with no height figure and the smallest size. An hour of speech arrives in the tens of megabytes, which sits comfortably inside the 150 MB single-file limit on free use. Your own software then turns the sound into text at whatever accuracy it manages.

Timings, speakers and the shape of the finished text

People mean different things by the word transcript, and the shape you need changes the tool you should reach for.

Settle which of these you want before you start, because converting between them afterwards is manual work.

• Running prose with no clock references, for reading or quoting.
• Timed lines that a video player can display, which is a subtitle file rather than a transcript.
• Text with speaker labels attached, which needs either a careful human pass or software built for it.
• A rough searchable dump, where machine errors matter less because you are looking for one phrase.

Quoting somebody accurately, and their rights in the words

A machine transcript of a technical talk misspells the terms that matter most. Check any sentence against the audio before you publish it as a quotation, because the mistakes cluster exactly where names and figures appear.

The speech is also somebody's work. Republishing an entire talk as text is a rights question in the same way that reposting the video would be. Vidzilla is not connected to YouTube and cannot grant permission over what a channel published.

Keep this tab open until the browser finishes writing the file for youtube transcript downloader. A prepared file only waits a short time before it expires from the temporary store.

Free use on Vidzilla covers three downloads a day and files up to 150 MB on video tools unless a converter page states its own quota. Pro raises those limits and removes advertising where ads appear.

Related pages

• Caption tracks on a YouTube video
• Taking the sound of a video on its own
• Changing a timed text file between formats
• Saving a long spoken-word recording