Video Transcript & Subtitle Extractor

What is Video Transcript & Subtitle Extractor?

Extract transcripts, subtitles and captions from public video URLs across 1,000+ platforms, including YouTube, Vimeo, TikTok, Instagram, X, Facebook, Twitch, TED, Bilibili, SoundCloud, and Reddit. Each accepted URL produces one dataset record with the original URL, inferred platform, transcript content, and useful metadata such as title, channel, duration, language, transcript source, segment count, character count, and subtitle file keys when available. YouTube playlist and channel URLs are expanded into individual videos. Native captions come first; Whisper AI transcribes videos without captions when you opt in. This Actor is usable through Apify and Apify MCP, and it returns one record per video.

Cost at a glance: one Transcript extracted event per video with a transcript, plus Whisper minute events only for platform-funded Whisper audio (none when you bring your own OpenAI key), plus Apify platform usage. Only videos that return a transcript are billed as transcripts. See the Pricing tab for current prices.

Best fit and connected workflows

This Actor fits workflows that start with a known public video URL and need structured transcript output. It routes naturally into:

• transcript indexing for search, embeddings, and knowledge bases
• content repurposing from talks, interviews, webinars, and podcasts
• accessibility file generation with SRT or VTT
• multilingual caption review for YouTube videos with captions
• AI agents that need one transcript record per input URL through Apify MCP
• metadata plus transcript extraction in one dataset row

Practical scenario

Maya, a content analyst, has three public video URLs from YouTube and Vimeo. She wants the title, channel, duration, transcript source, language, and transcript text for each one. She runs this Actor with outputFormat: "json" and includeMetadata: true. In the dataset, she reviews the transcript segments, sees whether each record came from manual captions, auto captions, translated captions, or Whisper fallback, and then decides which videos to turn into an internal summary and which ones to archive with subtitle files.

Input

Field	Type	Description
videoUrls (required)	array of objects	Public video URLs. Each item is one video, or a YouTube playlist/channel URL that is expanded into videos.
maxVideosPerPlaylist	integer	For YouTube playlist or channel URLs, the number of newest listed videos to process. Default 20; 0 disables expansion.
preferredLanguages	array of strings	ISO 639-1 codes in priority order. The first available match is returned. Use auto to accept any language.
outputFormat	string	Transcript format in each result: json, srt, vtt, text, or markdown.
useWhisperFallback	boolean	Optional fallback when native captions are unavailable. Defaults to false; enable it only with an OpenAI key or when Whisper coverage and its cost are explicitly desired.
whisperModel	string	auto selects a supported managed Whisper route; explicit whisper-1 keeps OpenAI.
openaiApiKey	string (secret)	Optional BYOK OpenAI API key for Whisper fallback billing through your OpenAI account.
translateTo	string	Optional ISO 639-1 code for YouTube caption translation.
includeAutoCaptions	boolean	Accept YouTube auto-generated captions.
maxDurationMinutes	integer	Set the maximum video length sent to Whisper. Native captions from longer videos are still extracted.
perVideoTimeoutSecs	integer	Time limit for one video (default 900 s). A slower video is skipped with a video_timeout note and the run moves on.
includeMetadata	boolean	Include title, channel, channel link and thumbnail in each result.
includeFullMetadata	boolean	Also return view count, like count, upload date, description and exact duration. Uses the slower full extraction route. Default off.
proxyConfiguration	object	Optional proxy settings.

Focused JSON input example

{
  "videoUrls": [
    {
      "url": "https://www.youtube.com/watch?v=dQw4w9WgXcQ"
    }
  ],
  "preferredLanguages": [
    "en",
    "es"
  ],
  "outputFormat": "json",
  "useWhisperFallback": false,
  "whisperModel": "auto",
  "openaiApiKey": "",
  "translateTo": "",
  "includeAutoCaptions": true,
  "maxDurationMinutes": 120,
  "perVideoTimeoutSecs": 900,
  "maxVideosPerPlaylist": 20,
  "includeMetadata": true,
  "proxyConfiguration": {
    "useApifyProxy": false
  }
}

Output

Each successfully extracted video produces one validated dataset record. Raw SRT and VTT files are also saved to the default key-value store. The output schema also exposes terminal run records for summary and downstream automation.

Field	Type	Description
videoUrl	string	Original URL processed.
platform	string	Platform inferred from the URL.
title	string or null	Video title from platform metadata.
channel	string or null	Uploader or channel name.
channelUrl	string or null	URL of the uploader or channel page.
durationSeconds	number or null	Video duration in seconds.
viewCount	number or null	View count at extraction time.
likeCount	number or null	Like count at extraction time.
uploadDate	string or null	Upload date in YYYYMMDD format.
thumbnail	string or null	Thumbnail image URL.
description	string or null	Video description, truncated to 1000 characters.
transcript	array or string	Transcript in the requested format. JSON returns {start, end, text} segments.
transcriptFormat	string or null	Format used for the transcript field.
transcriptSource	string or null	Source of the transcript: manual, auto, translated, or whisper.
language	string or null	ISO 639-1 code of the returned transcript.
isAutoGenerated	boolean or null	Indicates whether captions were auto-generated or Whisper-transcribed.
segmentCount	number or null	Number of timestamped segments in the transcript.
characterCount	number or null	Total transcript character count.
whisperMinutesCharged	number or null	Confirmed units at delivery; null while owner speech billing is pending. See final terminal receipts below.
srtKey	string or null	Key for the saved SRT file in the default key-value store.
vttKey	string or null	Key for the saved VTT file in the default key-value store.
playlistUrl	string or null	Playlist or channel URL the video was expanded from, when the input was a YouTube playlist or channel.
diagnostics	string or null	Issue detail recorded for a specific URL.

Sample output record (real run, transcript shortened)

The input was the TED channel URL with maxVideosPerPlaylist: 2, so the row carries playlistUrl.

{
  "videoUrl": "https://www.youtube.com/watch?v=ohpIw6iGXNA",
  "platform": "youtube",
  "title": "Why I’m Building Libraries Inside Prisons | Reginald Dwayne Betts | TED",
  "channel": "TED",
  "channelUrl": "https://www.youtube.com/@TED",
  "durationSeconds": 749,
  "viewCount": 41793,
  "likeCount": 175,
  "uploadDate": "20261015",
  "thumbnail": "https://i.ytimg.com/vi_webp/ohpIw6iGXNA/maxresdefault.webp",
  "transcript": [
    {
      "start": 4.201,
      "end": 7.538,
      "text": "I ate my first mango in a hole."
    },
    {
      "start": 8.506,
      "end": 10.307,
      "text": "What some called a shoe,"
    }
  ],
  "transcriptFormat": "json",
  "transcriptSource": "manual",
  "language": "en",
  "isAutoGenerated": false,
  "segmentCount": 215,
  "characterCount": 8323,
  "whisperMinutesCharged": 0,
  "srtKey": "srt-youtube-ohpIw6iGXNA.srt",
  "vttKey": null,
  "playlistUrl": "https://www.youtube.com/@TED"
}

How it works

The Actor expands YouTube playlist and channel URLs into individual videos and fetches YouTube captions through a fast route first (title, channel and thumbnail included; transcriptSource is manual, auto or caption when the source track cannot be told apart). If the fast route has no answer, or when you request translation or full metadata, it probes each video with the full extraction and prefers native captions first. If a probe is refused or times out, it retries once on a fresh residential session and then directly before recording the video in FAILURES. When multiple preferred languages are listed, it uses the first available match from the ordered list. When captions are available, it returns a validated transcript record and saves raw SRT and VTT files to the default key-value store—even when the video is longer than maxDurationMinutes. That cap applies only when no native captions are usable and Whisper fallback would incur per-minute transcription cost. The run also persists terminal outcome records such as COMPLETE, PARTIAL, CONFIG_ERROR, or UPSTREAM_FAILED, and stores per-URL diagnostics (for example no_transcript_available, invalid_video_url, video_timeout, probe_failed) in the FAILURES key-value record rather than the result dataset. Each video has its own time limit (perVideoTimeoutSecs), and the run stops cleanly before the platform timeout with a PARTIAL outcome that keeps every transcript already saved.

What's new (November 2026)

• Much faster YouTube runs. Captions now come from a fast route first: 12 videos finished in about 30 seconds (the previous full extraction took 30-60 seconds per video). Playlists and channels are fetched in parallel, and the full extraction still runs automatically if the fast route has no answer.
• New includeFullMetadata option for views, likes, upload date, description and exact duration.
• Cost example: 100 YouTube videos with captions = 100 × Transcript extracted + 1 × Actor start. Videos without a transcript are not charged. Store discounts lower the per-video price on higher Apify plans.
• Free plan: speech fallback is limited to videos up to 10 minutes.

Pricing

This Actor uses Apify Pay per event plus Apify platform usage. Open the live Pricing tab on the Actor page for the current charges and platform usage details.

Job	Events charged
10 videos, all with native captions	10 × Transcript extracted + 1 × Actor start
100 videos, 10 of them need Whisper (10 min each), platform-funded	100 × Transcript extracted + 100 × Whisper minute + 1 × Actor start
Same job with your own openaiApiKey	100 × Transcript extracted + 1 × Actor start (Whisper is billed by OpenAI to your key)
A playlist URL with maxVideosPerPlaylist: 20	up to 20 × Transcript extracted (only videos with a transcript)

Videos that end without a transcript (no captions, private, removed, refused by the site or timed out) are not charged as Transcript extracted. Apify platform usage is charged on top of the events; check the Pricing tab for current prices before large runs.

Use with AI agents (MCP)

This Actor is available as an Apify Actor usable through Apify MCP. The exact Actor identity is khadinakbar/video-subtitle-extractor.

Tool description: extract subtitles, captions, and AI transcripts from a supplied public video URL, then return one structured dataset record per video with transcript text, metadata, transcript source, and subtitle file references when available.

Extract transcripts from these public video URLs and return one dataset row per URL. Prefer native captions first, then Whisper fallback when needed. Include transcript source, language, and subtitle file keys in the output.

Output interpretation:

• transcript is the primary content field.
• transcriptSource shows whether the record came from manual captions, auto captions, translated captions, or Whisper.
• segmentCount and characterCount help estimate transcript size.
• srtKey and vttKey point to subtitle files in the default key-value store.
• diagnostics is available when a URL produces a record with an issue detail.

Provenance and scope:

• The dataset contains one record per processed video URL.
• The transcript record keeps the original videoUrl and inferred platform.
• Raw subtitle files are stored separately in the key-value store.
• The output schema also exposes run summary records for terminal outcome handling.

Pagination and cost guidance:

• For multiple videos, pass each video as its own URL, or pass a YouTube playlist/channel URL with maxVideosPerPlaylist.
• The primary event is Transcript extracted.
• Whisper-minute billing applies only when platform-funded Whisper transcription is used.
• Check the live Pricing tab for the current Pay per event and platform usage details before large runs.

JavaScript API example

const { ApifyClient } = require("apify-client");

const client = new ApifyClient({
  token: process.env.APIFY_TOKEN,
});

(async () => {
  const run = await client.actor("khadinakbar/video-subtitle-extractor").call({
    videoUrls: [
      { url: "https://www.youtube.com/watch?v=dQw4w9WgXcQ" }
    ],
    preferredLanguages: ["en"],
    outputFormat: "json",
    useWhisperFallback: false,
    includeMetadata: true
  });

  const dataset = client.dataset(run.defaultDatasetId);
  const { items } = await dataset.listItems();

  console.log("First record:", items[0]);
})();

Best results and outcome guidance

Use one public URL per item. Put preferred caption languages in priority order so the Actor can return the first matching transcript. Keep outputFormat: "json" when you need segment-level timestamps for AI processing, and switch to srt or vtt when you need subtitle files for editing tools. Leave includeMetadata on when you want title and channel context alongside the transcript; turn on includeFullMetadata only when you need views, likes, upload date or description. Use useWhisperFallback: true when you want coverage for videos without native captions and have supplied openaiApiKey (or explicitly accept the platform Whisper cost). The default false keeps prefilled and native-caption runs credential-free; set maxDurationMinutes to control Whisper cost for longer videos.

Continue the workflow

• Then use Vimeo Video Downloader to extend Video Transcript & Subtitle Extractor with a neighboring creator and media research source when the brief calls for Vimeo data.
• Use YouTube Transcript and Subtitle Data Scraper when every URL is a YouTube video and you only need YouTube caption data.
• Then use YouTube Comments Scraper to add audience comments and replies for the same YouTube videos.

Design note

I found that the live output contract centers on a single transcript record per video URL, with videoUrl and platform required in every dataset item. That makes the dataset shape straightforward for batch processing and agent workflows.

FAQ

Can I pass a playlist or channel URL?

Yes, for YouTube. A youtube.com/playlist?list=… URL or a channel URL (youtube.com/@handle, /channel/…, or its /videos or /shorts tab) is expanded into up to maxVideosPerPlaylist newest videos (default 20). Each output row carries playlistUrl. For other sites, pass individual video URLs.

Why is a video missing from the results, and was I charged?

Open the FAILURES key-value record. Each entry names the video and a reason such as no_transcript_available, no_captions_and_no_whisper_key, invalid_video_url, probe_failed or video_timeout. Videos without a transcript are not charged as Transcript extracted.

What happens with very long videos?

Native captions are extracted at any length. Whisper is only used up to maxDurationMinutes, and each video has a perVideoTimeoutSecs limit so the run keeps moving past one slow video.

Which transcript sources can appear in the dataset?

The dataset contract includes manual, auto, translated, and whisper as transcript sources.

What file formats are saved outside the dataset?

Raw SRT and VTT files are saved to the default key-value store, and the output schema exposes links for subtitle files and run summary records.

Can I use it with Apify MCP?

Yes. This Actor is built as an Apify Actor usable through Apify MCP, and the exact Actor identity is khadinakbar/video-subtitle-extractor.

How does it handle multiple caption languages?

Use preferredLanguages as a priority list. The Actor falls through the ordered list and returns the first available match.

Responsible use

Use this Actor only with public video URLs and in ways that respect the source platform's terms, applicable rights, and local law. Review the transcript source and metadata before reusing content, especially when captions are auto-generated or when Whisper fallback is used.

Speech billing receipts

Native captions and your own OpenAI key report whisperMinutesCharged: 0. Owner-funded speech results report whisperMinutesCharged: null while post-delivery billing is pending, plus whisperMinutesRequested and whisperBillingReceiptId. The final OUTPUT and RUN_SUMMARY records contain whisperBilling.receipts: requested units, confirmed accepted units, completion, and any uncertainty. A charge limit or lost billing reply produces a partial outcome and stops further work; an uncertain unit is never treated as accepted or retried automatically. Result and speech prices are unchanged.

Requested caption translations

An existing manual caption track in the target language is preferred. Native YouTube translation keeps the original caption source and respects your preferred languages. Turning off automatic captions excludes speech-generated sources; a translation derived from manual captions remains a manual source.

When a requested YouTube translation fails, useAiTranslationFallback (default true) can translate complete available captions using OpenAI. Only caption text is sent; timings stay unchanged. translationMethod distinguishes native and ai. Set this option to false for native-only translation. No extra translation event or speech-minute fee is added; existing transcript and platform-usage pricing stays unchanged.

The AI route requires a verified creator-cost allowance and a verifiable target language. Long or otherwise unaffordable translations stop before the paid request. Partial, refused, wrong-language or uncertain results are not delivered or billed as transcripts. Provider failures may still incur a bounded creator cost; this route does not promise every language or length can fit the current price.

Automatic speech and exact model selection

Speech remains optional. Native captions are tried first. whisperModel: "auto" selects a supported multilingual Whisper route when you enable useWhisperFallback. Explicit whisper-1 selects OpenAI and is never silently replaced. With your own OpenAI key, auto also resolves to whisper-1; your key is sent only to OpenAI. speechModel records the configured model used.

The existing transcript and managed speech-minute prices are unchanged. No managed speech-minute event is charged with your own OpenAI key. Audio duration is checked from the downloaded file, and complete timed segments are required before transcript delivery. If the selected managed model cannot serve the video's length at the current service price, choose Automatic speech or supply your own OpenAI key. Quota exhaustion, incomplete responses and uncertain attempts stop without a transcript result fee; requests are not automatically replayed. Files must fit the 25 MB upload limit. The managed Large V3 route supports up to 120 minutes per file, subject to available provider quota.

YouTube transfer limits

For each YouTube video using the Actor's configured connection routes, metadata, captions and retries share a transfer allowance. Native captions and caption translation share 8 MiB; managed speech uses 16 MiB for a maximum duration of three minutes or less, and 32 MiB for longer duration limits. A video that reaches the allowance stops with a failed-video diagnostic and no transcript result fee. Duration and file-size limits still apply separately.

Playlist/channel listing has a separate shared 8 MiB allowance across its discovery routes and retries. This discovery cost belongs to the collection and is not counted as a new revenue allowance for each child video. If listing reaches its allowance or cannot save its transfer checkpoint, its child videos are not admitted.

These guards preserve existing paths for a caller-supplied custom proxy or OpenAI key, which remain outside the transfer allowances above. Other video sites also retain their existing paths. Transfer counters describe observed traffic; they do not guarantee an invoice amount. Existing prices and included platform usage are unchanged.

When the current run confirms that platform usage is paid by the caller, managed YouTube extraction tries direct access first and can fall back to Apify's configured proxy. This also applies to playlist discovery. A caller's custom proxy or OpenAI key preserves its existing connection path. If the run's usage payer or platform share cannot be verified, managed YouTube work stops with CONFIG_ERROR before discovery, proxy allocation or video requests, with no transcript or managed speech-minute fee. Other video sites in a mixed request still run and their completed results are retained. Connection choice can change the total platform usage consumed; per-event prices and the included-usage setting stay unchanged.

translateTo still applies to YouTube captions, not to speech fallback. Native SRT/VTT/JSON behavior, output formats and timing preservation are unchanged.