How to Get Subtitles From Video (or a Clean Transcript)

2026-08-13

There are two completely different ways to get subtitles from a video, and picking the wrong one wastes an afternoon. If the video already carries a subtitle track, you can pull it out in seconds — no speech recognition involved. If it doesn't, the text has to be created from the audio, either by a human typing along or by automatic speech recognition. And before you start, it's worth asking a question most people skip: do you actually need subtitles — timed lines that sit on top of the picture — or just the text of what was said? This guide covers both routes and helps you choose.

Two ways to get subtitles from video: extract or recognize

"Getting the subtitles" means something different depending on how the video was made. Three situations cover almost every case:

• A separate subtitle track exists. MKV files, DVDs, and many streaming downloads carry "soft" subtitles as an embedded track. Tools like ffmpeg or a video player's export function can extract subtitles from video files like these directly — the text is already there, character for character.
• Captions exist on the platform. YouTube and most video platforms store captions alongside the video, whether uploaded by the creator or auto-generated. Creators can download their own caption files from the platform's dashboard.
• The subtitles are burned into the picture — or there are none. Hardcoded subs are just pixels; there is no text to extract. Same story for a plain recording with no captions at all. In both cases, the only way to get the text is to recognize the speech itself.

The first two cases are extraction problems and take minutes. The third is a transcription problem — and it's the one people actually get stuck on.

Auto-generated vs manual subtitles: an honest comparison

Manual subtitling — a person listening and typing timed lines — produces the best result and always will. Humans catch mumbled names, sarcasm that changes punctuation, and the difference between two similar-sounding product names. They are also slow: careful subtitling takes several times the length of the video.

Automatic recognition flips the trade. A modern speech-to-text engine handles an hour of clear speech in minutes and gets the vast majority of words right, but it will stumble on heavy background noise, people talking over each other, and unusual proper nouns. The practical answer for most work is neither extreme:

1. Generate subtitles automatically to do the 95% of typing no human should do by hand.
2. Edit the output — fix names, terms, and the handful of misheard phrases.
3. Reserve fully manual work for cases where errors are costly: broadcast, legal, medical, or accessibility-critical publishing.

For an internal training video or a lecture you need to review, edited auto-generated text is entirely adequate. For the subtitles on your product's launch film, budget for a human pass.

SRT in two minutes

If you do need proper subtitles, you'll almost certainly meet the SRT format. It's a plain text file, readable in any editor, made of numbered blocks:

1
00:00:01,000 --> 00:00:04,200
Welcome back. Today we're looking
at the quarterly numbers.

2
00:00:04,400 --> 00:00:07,000
Let's start with revenue.

Each block has a sequence number, a start and end time, and one or two lines of text. Conventions worth knowing: keep lines short (roughly 40 characters), never more than two lines per block, and leave each block on screen long enough to read — about 1 to 7 seconds. WebVTT (.vtt) is the web-native cousin of SRT with nearly identical structure; most platforms accept either. Because both are plain text, converting video captions to text is trivial: strip the numbers and timestamps and you have a readable script.

When a transcript beats subtitles

Here's the part that saves people the most time: a large share of "I need the subtitles" requests are really "I need what was said." If your goal is any of the following, timed subtitle blocks are the wrong container:

• Quoting or writing. Pulling quotes for an article, a report, or show notes.
• Summarizing. Turning a 40-minute recording into a one-page brief.
• Searching. Finding the moment a specific decision was made.
• Translating. Getting the content into another language for readers, not viewers.

Subtitle files chop sentences into fragments to fit the screen; a transcript keeps them whole. This is exactly where a tool like Oratext fits: upload the video file as is — no need to extract the audio track yourself — and get a full transcript, with speech recognized in around 99 languages and the language detected automatically. From the same transcript you can then remove filler words, generate a summary, or translate the text in one click. The broader workflow is covered in our guide to converting video to text, and if the end goal is a foreign-language version, see audio translation.

A practical decision path

To pull it all together, here's the short version:

1. Video has a soft subtitle track or platform captions? Extract or download them — done.
2. Need the content as readable text? Transcribe the video and work with the transcript: clean it, summarize it, translate it.
3. Need on-screen subtitles and none exist? Generate the text automatically, edit it, then time it in a subtitle editor or let your video platform's caption tool align it.
4. Publishing something high-stakes? Add a human review pass before it goes live.

FAQ

Can I get the text of a video from my phone?

Yes. Send the video file to the @oratextbot bot in Telegram — it handles files up to 20 MB and returns the transcript right in the chat, along with cleanup, summary, and translation options. For longer videos, use the web version at oratext.com.

Does this work for videos in other languages?

Oratext recognizes speech in around 99 languages and detects the spoken language automatically, so you can transcribe a video in one language and translate the result into another without any extra tools.

Can subtitles burned into the video be extracted as text?

Not directly — hardcoded subtitles are part of the image, not a text track. The practical route is to ignore them and transcribe the audio instead, which usually produces cleaner text anyway.

Need the text of a video today? Try it free at oratext.com — you get 3 minutes of transcription every day, no registration required — or send the file to @oratextbot in Telegram. Paid plans start at about 60 Telegram Stars (roughly a dollar) per month.