Video to Text: Transcribe Videos and YouTube

Start by asking whether the video already has captions. If it does, copy the text and skip transcription. If it does not, extract the audio track and run it through a speech recognition tool, then check the draft against the video, because these tools make errors and occasionally invent whole phrases ¹. Downloading a video from YouTube to do this is restricted by YouTube's terms, and the copyright question depends on what you do with the result. The sections below cover each route and the limits we could confirm from primary sources.

If a video already has captions, copy its text before you transcribe anything

A caption track is text already timed to the speech in a video, so text from a caption track saves the transcription step. The catch is accuracy. The text was produced by a machine or a person, and it has its own errors. In a 2017 study of YouTube's automatic captions, the authors found lower accuracy for women and for speakers from Scotland ². Read the copied text against the audio before you rely on it, especially for names, numbers and quotations.

Copying is the right first step when you only need the words and the platform already has them. It is the wrong step when you need a verified record, such as a legal or research transcript, because the source text has not been checked by you. For those uses, treat the copied text as a draft, and follow the proofreading steps in our guide to transcribing audio.

YouTube's terms allow viewing its content but bar downloading it

YouTube's terms let you view or listen to its content for your personal, non-commercial use ³. The same section says you are not allowed to download any part of the service or its content, except where the service specifically permits it ⁴.

In our reading of the terms, copying a transcript that the platform shows you falls under the permission to view. Saving the video file, or running a third-party downloader to get its audio, is the step the terms restrict. If you need the transcript from a YouTube link, see how to get a YouTube transcript. If a video has no transcript to show, why the YouTube transcript is not showing covers the common reasons.

We could not open YouTube's own help pages while researching this guide, so we have left out the menu steps for finding a transcript. Those steps change, and the help pages are the current reference.

Five steps turn a video with no captions into a checked transcript

Use the same five steps you would use for any recording. They do not depend on the video platform, and they are the steps the rest of this guide assumes.

1. Get the audio. Use the file you own or have permission to use. If you only have a link, see the copyright and terms section above before you go further.
2. Set the language. Choose the spoken language yourself when you know it. Otherwise the tool may guess.
3. Transcribe. Run the audio through your model, on your own computer or in the cloud.
4. Check the draft against the video. Play the video at the points where the draft looks wrong, and check names, numbers and any text the speaker reads out.
5. Add timing if you need captions. A draft with no times is not a caption file. Use the SRT or WebVTT step covered in the final section.

Video editors can transcribe footage too, but the features differ by product

Some video editors have a transcription tool built into the timeline or media panel. What each tool costs, which languages it covers, and whether it runs on your own computer differ by product and version. We did not verify the current features of any editor for this guide, so this guide does not state them. Check the help page for the editor you use before you plan a workflow around its transcription tool.

Product-specific answers are on our question pages for transcribing audio in Premiere Pro and transcribing audio in DaVinci Resolve.

Without captions, transcribe the video's audio track like any other recording

A video with no captions is an audio recording with pictures. Pull out the audio track, then run it through a speech recognition model, the same way you would transcribe any automatic speech recognition file. Whisper, the open model from OpenAI, supports 98 languages, and its accuracy varies by language ⁵. The Python package needs the ffmpeg command-line tool installed on your system, ⁶.

Model size sets the trade-off between speed and accuracy. Relative to the large model, the turbo model runs at about 8 times its speed ⁷, the base model about 7 times ⁸, the small model about 4 times ⁹ and the medium model about 2 times ¹⁰.

Speed is not the same as accuracy, and the accuracy numbers come from different test sets. In a 2024 study, Whisper large-v2 had an average word error rate of 2.9 percent on English lecture recordings ¹¹. On the Open ASR Leaderboard's English short-form tests, Whisper Large v3 Turbo averaged 7.83 percent ¹². A lecture recording and a set of short clips are different material, so neither number predicts the error rate on your video. Test the tool on a minute of your own video first.

Time on a real recording depends on the implementation as well as the model. In one benchmark of a 13-minute recording ¹³, openai/whisper took 2 minutes 23 seconds with the large-v2 model on a GPU in fp16 precision ¹⁴. The faster-whisper project's own benchmark, run on the same GPU, took 59 seconds with int8 precision ¹⁵. Its authors published the figures, so treat them as their result rather than an independent test. Running faster-whisper on a GPU also needs NVIDIA libraries installed first ¹⁶.

Memory matters as much as speed on a home computer. The turbo model needs about 6 GB of video memory, against 10 GB for the large model ^{17 18}, so it is the lighter choice on a GPU with limited memory. When whisper.cpp runs with Core ML on an Apple device, the first run is slow, because the Neural Engine service compiles the model to a device-specific format ¹⁹. Expect the first run to take longer than later ones.

Review the whole draft, because a model can invent text. In a 2024 study, roughly 1 percent of Whisper transcriptions contained entire hallucinated phrases or sentences that did not exist in the audio ¹. The same study tested Whisper as it stood in 2023 ²⁰, so the figure describes that version, not every later one. A 2025 study found that when non-speech sounds were played to Whisper, a set of hallucinated phrases appeared frequently ²¹, so check quiet sections and music first.

Set the language and vocabulary before you run the file

If you leave the language unset in the openai-whisper package, the program detects the spoken language itself ²². Setting the language yourself removes one source of error when the video is mostly in one language. Names are a separate problem. Whisper's initial prompt can carry custom vocabulary or proper nouns, which makes those words more likely to be transcribed correctly ²³. Put a short list of the names and terms in the video into that prompt before you run it.

A checklist for reviewing the draft

For a step-by-step workflow for any file, see how to transcribe audio to text. For model choices and local setup, see transcribing audio with Whisper, ChatGPT or Gemini.

Audio playing on your computer can be transcribed live, but the text is not always saved

If you cannot get a file, a live caption feature can transcribe the sound as it plays. Each platform has one, and they differ in what they need and what they process.

On a Mac, you can switch between captioning the computer's audio and the microphone, which helps with an in-person conversation ²⁴. Apple says the accuracy of Live Captions on Mac may vary and should not be relied on in high-risk or emergency situations ³¹, so treat it as a convenience. For more on live captions, see live captions on phones and tablets and the Mac guides.

Live captions show text as the audio plays. Whether you can save that text, and in what format, depends on the app. Check for a copy or export option before you rely on the output. If you need a saved file, a recording or a downloaded file is the better input than a live caption.

Downloading someone else's video to transcribe it raises copyright and terms questions

Two separate questions apply. The first is the platform's terms, covered above: YouTube's terms restrict downloading its content unless the service permits it ⁴. The second is copyright, and the answer depends on the country and the use.

In the United States, section 107 of the Copyright Act sets out the fair use exception. Its first factor is the purpose and character of the use, including whether the use is commercial or for nonprofit educational purposes ³². The section lists purposes such as criticism, comment, news reporting, teaching, scholarship and research as examples of fair use ³³. The statute lists factors rather than a fixed test, and courts decide each case on its facts. A transcript of a whole video for your own convenience is not one of the listed purposes, but that alone does not settle the question either.

This section describes US law only, and it is general information, not legal advice. If you need a transcript of someone else's video for work, the safer routes are to ask the rights holder, to use captions the platform already provides, or to get permission. If the video is your own recording, the copyright question does not arise in the same way.

A transcript becomes captions only once each line is timed to the video

A transcript is text in order. Captions are text plus timing, so each line appears on screen when the speech happens. The two common caption formats store that timing. A SubRip subtitle file (SRT) lists each line with its start and end time, and a WebVTT file does the same for web players. A plain transcript can be turned into captions, but the timing has to be added, either by a tool that aligns the text to the audio or by hand.

The timing is also what accessibility rules check. WCAG success criterion 1.2.2 asks for synchronized captions on non-live video, and it is a Level A criterion ³⁴. A plain transcript on its own does not do that job. For what to add to a transcript for captions, including closed captions and timestamps, see captions and subtitles. If you also need to label who is speaking, see timestamps and speaker labels and the term speaker diarization.

Choose the route by what the video has and what you need

For a recording you already have, the free audio to text tool transcribes files in your browser. Our transcription software guide explains how to compare tools by accuracy, privacy and cost. Yap, which we make, handles dictation on your own device for Mac and Windows, and you can download the app.

For the wider picture, start with transcription, then read meeting transcription, voice memos to text and human vs AI transcription, which covers accuracy, cost and speed on the same terms as this guide.