All guides

How to turn YouTube captions into a clean transcript

By · Updated

In short: for your own videos, download the captions from YouTube Studio as a .vtt, .srt or .sbv file. For any video that has captions, the transcript panel lets you copy the text. Either way, the raw result needs cleaning — timestamps, word-level timing codes and lines repeated two or three times — before it reads like a document.

Two ways to get the captions

From YouTube Studio, for your own videos

If you uploaded the video, YouTube lets you download every caption track it has, including the automatic one:

  1. Open YouTube Studio and choose Subtitles in the left-hand menu.
  2. Select the video, then choose Edit beside the language under Subtitles.
  3. In the caption editor, open Options → Download subtitles. Use the format offered by Studio; delsub accepts .vtt, .srt and .sbv.

These steps follow YouTube’s current help documentation; account state, device and interface updates can change what you see. Any of the three formats works; the formats guide explains the differences.

From the transcript panel, for any video with captions

On a video page, YouTube can show the captions as a scrolling transcript:

  1. Expand the video description and click Show transcript.
  2. The transcript opens beside the video, one caption per line, each with its timestamp.
  3. In the panel’s menu, choose Toggle timestamps to hide them.
  4. Select the text in the panel and copy it.

What you get is one short caption per line, without timing. It is readable, but broken into fragments of five or six words, often without punctuation.

What is wrong with the raw captions

An automatic caption file from YouTube is designed for a player, not for a reader. This is the start of a typical one:

WEBVTT
Kind: captions
Language: en

00:00:00.000 --> 00:00:02.330 align:start position:0%

so<00:00:00.320><c> today</c><00:00:00.640><c> we're</c><00:00:00.960><c> going</c><00:00:01.200><c> to</c><00:00:01.440><c> talk</c>

00:00:02.330 --> 00:00:02.340 align:start position:0%
so today we're going to talk

00:00:02.340 --> 00:00:05.150 align:start position:0%
so today we're going to talk
about<00:00:02.640><c> subtitles</c><00:00:03.120><c> and</c><00:00:03.360><c> how</c><00:00:03.600><c> they</c><00:00:03.840><c> work</c>
[Music]

Four separate problems are mixed together here:

  • Header and timing lines — WEBVTT, Kind:, Language: and valid pairs of start/end timestamps, including their positioning settings. A spoken expression such as A --> B is kept.
  • Word-level timing — the <00:00:00.320> codes and <c> tags around each word, which make the words light up one by one in the player.
  • Repeated lines — “so today we're going to talk” appears three times. Automatic captions scroll up line by line, and each cue repeats the previous line to keep it on screen. Why auto-generated captions repeat every line explains the mechanism.
  • Sound annotations — recognized cues such as [Music], [MUSIC PLAYING] and [Applause]. Standalone music notes and notes around a full lyric line are removed without deleting the sung words.

Cleaning it in one step

Drop the file into delsub (or paste the text) with the default options and those four problems are handled together. The example above comes out as:

so today we're going to talk about subtitles and how they work

Each rule is a separate switch, so you can keep what you need — the timestamps for a reference, say, or the sound cues for an accessibility review. The two that change the result the most are:

  • Roll-up repeats. Removes a line that repeats the previous one, or that the next line extends. Turn it off if the speaker genuinely repeats themselves and you want to keep it.
  • Rebuild paragraphs. Joins caption lines into sentences and starts a new paragraph after a line that ends with closing punctuation, such as a period, a question mark or an exclamation mark.

When automatic captions have no punctuation

This is the main limit of any transcript built from automatic captions. When the speech recognizer adds no periods, there is nowhere to break a paragraph, so “Rebuild paragraphs” produces one long block. You have two practical options:

  • Turn Rebuild paragraphs off to keep one caption per line, which is easier to scan and to edit by hand.
  • Keep the single block and add punctuation afterwards, by hand or with a writing tool. The text is already free of timing codes and repeats, which is what makes that step practical.

Captions written by a person — or uploaded by the creator — are usually punctuated, and paragraphs come out naturally. Keep in mind too that cleaning does not correct recognition errors: if the automatic captions misheard a word, the transcript will contain the same mistake.

Using someone else’s transcript

A video’s captions are part of its creator’s work. A transcript for your own notes, research or accessibility needs is a different thing from republishing it; for anything public, quote short passages with a link to the video, or ask the creator. For your own videos, the transcript is yours to reuse — as a blog post, show notes or the starting point for a summary.

Reproduce this exact cleanup

The VTT above is a synthetic example, not a transcript downloaded from a real video. Its input and output were checked with delsub’s decoder, parser and paragraph rebuilder on 21 October 2026.

  1. Download the complete YouTube-style VTT and load it into delsub.
  2. Turn all seven switches on. Options are remembered between visits, so verify them before comparing.
  3. Compare the result with the exact expected UTF-8 output. It ends with a newline; it does not invent punctuation or missing words.

The sample includes the words “they work” in its input. Cleanup removes structure and duplicates; it cannot reconstruct words absent from the source. The recorded results and file hashes describe the checked options. This test does not verify the current Studio interface or speech-recognition accuracy.

Sources and scope

Primary documentation checked on 21 October 2026. The examples above test delsub’s own behavior; they do not certify every subtitle file or third-party interface.

Have a caption file or a copied transcript? Paste it into delsub: the clean text appears as you type, and you can copy it or download it as .txt or .md.