Build a Transcript Dataset from a Batch of Lecture Videos

A folder of lecture links becomes a queryable text dataset in one run. Thirdwatch's YouTube Subtitle Scraper returns every caption track a video offers as timestamped segments plus full video metadata — no API key, no OAuth.

Skip the setup: Run this as a ready-to-go task on Apify — pre-loaded with the configuration from this guide.

Why build a lecture transcript dataset

Lecture series, conference talks and explainer channels are some of the highest-quality free educational content online — and none of it is searchable as text until the captions are extracted. Doing that by hand means opening each video and copy-pasting the transcript panel; doing it via the official API means OAuth and caption-authorisation walls.

A list of URLs in, a dataset of transcript rows out — that is the entire pipeline.

The subtitle scraper reads the public caption endpoints directly, so the batch path is a list of URLs in and transcript rows out.

How does this compare to the alternatives?

Why this Actor

• Manual and auto-generated tracks both supported — captionSource labels which you got.
• Every segment carries startMs, durationMs and text, so timestamp-level alignment survives.
• availableLanguages lists every track the video offers — discovery and extraction in one pass.
• Each video gets its own proxy session with a fresh-exit retry, so multi-video batches stay reliable.
• Full metadata on the row: title, channel, duration, view count, upload date.

How to do it in 3 steps

Step 1: Configure the input

Set the inputs as shown below — videoUrls takes the targets, languages picks the caption language(s) and preferManual chooses human-written over auto-generated tracks.

Step 2: Run the Actor

Run it from the console, the API, or the linked saved task. One dataset row is written per video.

Step 3: Use the output

Each row carries the video metadata, the caption track as segments (with startMs/durationMs/text), and the joined transcript string.

{
  "videoUrls": [
    "https://www.youtube.com/watch?v=dQw4w9WgXcQ",
    "https://www.youtube.com/watch?v=kJQP7kiw5Fk"
  ],
  "languages": [
    "en"
  ],
  "preferManual": true,
  "proxyCountry": "US"
}

Each dataset row looks like:

{
  "videoId": "dQw4w9WgXcQ",
  "url": "https://www.youtube.com/watch?v=dQw4w9WgXcQ",
  "title": "Rick Astley - Never Gonna Give You Up (Official Video)",
  "channel": "Rick Astley",
  "language": "en",
  "captionSource": "manual",
  "segmentCount": 61,
  "segments": [
    {
      "startMs": 18170,
      "durationMs": 2290,
      "text": "We're no strangers to love"
    }
  ],
  "transcript": "We're no strangers to love ..."
}

What to watch for

Heavily scraped videos sometimes return a 429 on the caption file even after the fresh-exit retry — those come back as error rows, not silent gaps. Videos with no captions at all behave the same way, which is itself a useful audit signal.

Related use cases

• Extract a YouTube Video's English Transcript as Structured Data
• Harvest YouTube Subtitles in Multiple Languages at Once
• Audit Which YouTube Videos Only Have Auto-Generated Captions
• Turn YouTube Podcast Episodes into a Searchable Transcript Corpus
• All Thirdwatch use-case guides

Run the YouTube Subtitle Scraper on Apify Store — pay per result, free to try, no credit card to test.