🎬 Kuantum Video Transcript Scraper — YouTube, Vimeo, TED & 1000+ Sites

Store support: Apify Issues. Unavailable captions, invalid inputs, and failed source requests are not successful transcript results and are not billed as result events.

Billing

Successful transcript results use the Actor's configured Apify pricing. Unavailable captions and failed inputs are not successful billable results.

Turn a captioned video into clean, structured, AI-ready text. This production API is live-canary tested on YouTube and TED. Every successful transcript ships as JSON and LLM-ready formats (clean Markdown + token-aware RAG chunks), so it drops straight into ChatGPT, Claude, a vector database, or your analytics pipeline.

No API keys. No monthly lock-in. Pay only for what you extract.

YouTube is the flagship and works great. TikTok, X (Twitter), Facebook, and Instagram are limited — see Supported platforms before you rely on them.

---

⚡ Why people choose this scraper

	This Actor	Typical transcript scrapers
LLM-ready RAG chunks (timestamped, token-sized)	✅ Built in	❌ Raw text only
Production-verified sources	✅ YouTube + TED	⚠️ Often unverified
Languages	✅ Any available + variant fallback	⚠️ Often English-only
YouTube reliability (JS-challenge / PoToken solved)	✅ Yes	⚠️ Breaks often
Pricing	✅ Pay-per-result, no subscription	❌ $19.99+/mo rental
Output formats	✅ JSON · text · Markdown · chunks	⚠️ One format

The edge in one line: this Actor delivers embedding-ready chunks and clean Markdown, not merely raw captions, with transparent billing only when a transcript is delivered.

🚀 What you can build with it

• A "chat with any YouTube channel" bot — transcribe a creator's back-catalog, embed the chunks, ship a RAG assistant.
• Competitor & content listening — bulk-pull YouTube, podcast, webinar, and conference-talk transcripts and run sentiment or topic analysis.
• Research datasets — thousands of talks/podcasts/interviews as clean, timestamped text in any language.
• Content repurposing — turn a video into a blog post, summary, or show notes with the Markdown output.
• Training / fine-tuning data — normalized transcripts across platforms, ready to feed a model.

🌍 Production support

The actor works only when a video exposes a caption/subtitle track. The underlying yt-dlp engine supports many extractors, but extractor support is not the same as production-verified transcript support.

✅ Production-verified — live daily canaries validate: YouTube (flagship, with residential proxy) · TED.

⚠️ Experimental — Vimeo and other yt-dlp-supported caption-bearing sites may work, but are not advertised as production support until their own live canaries pass. TikTok, X (Twitter), Facebook, and Instagram commonly require login/cookies or expose no caption track, so they are not reliable targets.

Extraction needs the video to expose a human or auto-generated caption track. Videos without captions create no dataset item and no transcript-result charge.

Only need YouTube? There's a focused YouTube Transcript API edition — same engine, YouTube-scoped listing.

📥 Input

Field	Type	Default	Description
videoUrls	array	— (required)	Video/audio URLs, or channel/playlist URLs (auto-expanded into their videos).
maxVideosPerList	integer	20	Cap per channel/playlist (max 200; total run ceiling 1,000).
language	string	en	Preferred subtitle language; falls back to closest variant (es → es-419).
includeAutoCaptions	boolean	true	Use auto-generated captions when no human track exists.
outputFormats	array	all	json, text, markdown, chunks.
chunkTokens	integer	400	Tokens per RAG chunk (a real, enforced cap — accurate for CJK too).
tokenizer	string	o200k	Tokenizer for sizing/counts: o200k (GPT-4o / embedding-3), cl100k (ada-002 / GPT-3.5-4), or chars (fast heuristic).
translateTo	string	—	Also return the transcript translated to this language (e.g. es). YouTube-only, best-effort.
includeTimestamps	boolean	true	Include per-segment start/end array.
bestEffort	boolean	false	Adds diagnostic logging for unavailable captions; no result row is created or billed.
maxConcurrency	integer	5	Videos processed in parallel.
priorityProcessing	boolean	false	Bulk priority add-on. Charged only on successful delivery after the premium event is published.
webhookUrl / webhookSecret	string	—	Send one HMAC-signed final-run summary to your endpoint.
proxyConfiguration	object	Residential	Residential is the reliability default; choose datacenter only for lower-cost experiments.

{
  "videoUrls": [
    "https://www.youtube.com/watch?v=dQw4w9WgXcQ",
    "https://www.ted.com/talks/sir_ken_robinson_do_schools_kill_creativity"
  ],
  "language": "en",
  "outputFormats": ["json", "text", "markdown", "chunks"],
  "proxyConfiguration": { "useApifyProxy": true, "apifyProxyGroups": ["RESIDENTIAL"] },
  "webhookUrl": "https://example.com/apify/transcript-finished",
  "webhookSecret": "replace-with-a-unique-16-character-minimum-secret"
}

📤 Output

One dataset item per URL. Real output from an example run — Steve Jobs' 2005 Stanford Commencement (input: {"videoUrls":["https://www.youtube.com/watch?v=UF8uR6Z6KLc"]}) → 341 timestamped segments, 7 RAG chunks with exact o200k token counts:

{
  "platform": "youtube",
  "title": "Steve Jobs' 2005 Stanford Commencement Address",
  "channel": "Stanford",
  "duration": 904,
  "view_count": 48703066,
  "upload_date": "2008-03-08",
  "available_languages": ["en", "es", "fr", "ja", "..."],
  "transcript_found": true,
  "selected_language": "en",

  "text": "[Music] this program is brought to you by Stanford University. I am honored to be with you today...",
  "transcript": [{ "text": "[Music]", "start": 1.57, "end": 8.96 }],
  "markdown": "# Steve Jobs' 2005 Stanford Commencement Address\n\n**Channel:** Stanford  \n**Duration:** 15:04\n\n---\n\n[Music] this program is brought to you by...",
  "chunks": [
    { "chunk_index": 0, "text": "[Music] this program is brought to you by Stanford University...", "start": 1.57, "end": 164.76, "token_estimate": 385 }
  ]
}

• text — full cleaned transcript.
• transcript — timestamped segments.
• markdown — metadata header + paragraphed body, ready to paste into an LLM.
• chunks — token-aware, timestamped chunks, ready to embed for RAG. Sizing and token_estimate use a real tokenizer (tiktoken, default o200k_base), so chunkTokens is an accurate, enforced cap even for Chinese/Japanese/Korean, emoji, and code (a chars/4 heuristic under-counts those ~2–3×). Counts are an accurate proxy for Claude/Gemini, which don't ship offline tokenizers.

💸 Pricing

Pay-per-result — you pay per video successfully transcribed, with no monthly subscription. Run one video or ten thousand; the cost scales with what you actually use. Free Apify tier available to try it.

Unavailable captions, extraction failures, and diagnostic logs do not produce a default-dataset item, so they do not trigger the transcript-delivered event. The bulk priority add-on is intentionally opt-in and must be published as the priority-transcript pay-per-event event before it is offered to customers.

🔔 Automation webhook

Set both webhookUrl and webhookSecret to receive exactly one final summary after a successful run. The payload is JSON and its X-Apify-Transcript-Signature header is sha256=<HMAC-SHA256(body, webhookSecret)>. X-Apify-Transcript-Delivery-Id is stable per run; store it at your receiver to make retries idempotent. Deliveries require public HTTPS, time out after 15 seconds, never follow redirects, and retry on network failures, HTTP 429, and HTTP 5xx (up to three attempts). A webhook failure never changes a successful Actor run into a failure.

{
  "actorId": "…",
  "runId": "…",
  "datasetId": "…",
  "status": "SUCCEEDED",
  "urlsRequested": 20,
  "transcriptsDelivered": 18,
  "unavailableOrSkipped": 2,
  "failed": 0,
  "priorityTranscriptsDelivered": 0,
  "finishedAt": "2026-08-14T00:00:00.000Z"
}

❓ FAQ

How do I get a YouTube transcript in bulk? Paste any number of YouTube URLs into videoUrls and run — you get JSON + Markdown + chunks for each.

What happens if I submit the same URL twice? Exact duplicate input URLs are deduplicated before extraction, so a transcript is not delivered or charged twice in one run.

Does it work without captions on the video? It reads existing subtitle tracks (human or auto-generated). If a video has none, it is skipped, logged, and never billed.

Some YouTube videos fail — why? YouTube can block shared datacenter IPs. Residential proxy is the default reliability setting; do not switch it to datacenter unless you have tested your workload.

Can I get the transcript translated? Set translateTo (e.g. es) and the result includes a translated object alongside the original. It uses YouTube's auto-translation, so it's YouTube-only and best-effort — not every video/language is offered; translated is null when it isn't. Adds a second extraction per video.

Can I transcribe a whole channel or playlist? Yes — paste a channel or playlist URL into videoUrls and it expands into that list's videos (most recent first), bounded by maxVideosPerList (default 20, max 200) and a 1,000-video run ceiling. A /watch?v=…&list=… link is treated as the single video, not the playlist.

Does it work for Vimeo or social platforms? Vimeo and other extractors remain experimental until separately verified. Social platforms commonly require login/cookies and are not a reliable use case today. Use this Actor for YouTube and TED.

Which languages are supported? Any language the source exposes. Ask for en, es, fr, pt, de, etc.; it falls back to the nearest variant automatically.

Do I need my own API key? No. Just add URLs and run.

Is this legal? It extracts publicly available subtitle tracks. You're responsible for respecting each platform's Terms of Service and applicable copyright/data-protection law — only process content you're permitted to access.

---

Keywords: video transcript scraper, YouTube transcript API, TED transcript, captions to JSON, transcript for LLM, RAG, video to text, bulk transcript, webhook.