YouTube Transcript Scraper — SRT, VTT, JSON & Plain Text with Whisper AI Fallback

Turn any YouTube video or Short into a clean transcript in one run. Every row carries the full text, timestamped segments, ready-made SRT and VTT subtitle files, and the video's core metadata (title, channel, views, duration, description). When a video has no captions at all, the optional Whisper AI fallback downloads the audio and transcribes it with speech-to-text — something caption-only scrapers simply return empty for.

How it works

✨ Why use this scraper?

• Several ways to the captions — when one way to a video's captions comes back empty, the next one is tried automatically, and opt-in Whisper AI speech-to-text comes last. A temporary error on one video never costs you the run.
• Whisper AI for caption-less videos — the differentiator: videos with no captions still come back transcribed (opt-in, billed as a separate premium event so you never pay it unknowingly).
• Every format in one row — plain text for LLM pipelines, timestamped JSON segments for analysis, SRT and VTT for subtitle workflows. No post-processing.
• Language selection with auto-translate — request en, de, es… and where the exact track is missing, YouTube's translated track is used when available.
• Metadata included free — title, channel and @handle, publish date, view count, duration, keywords, description and thumbnails ride along on every transcript row.
• Every caption track listed — availableCaptionTracks names each language YouTube has for the video and whether a person wrote it or it is auto-generated, so you can pick the track yourself.
• Live stream replays work — paste the youtube.com/live/… link as-is. A stream that is still live, or ended so recently that YouTube has not published its captions, is reported as live_not_ready and not charged.
• Bulk-friendly — paste hundreds of URLs, whole channels or playlists in one run; failed videos are isolated and never billed.
• Paste the input you already use — videoUrl + targetLanguage (pintostudio's input) and urls: [{ "url": … }] run unchanged, and every row also carries a data array of start / dur / text, the segment shape those pipelines read.
• Says which track it read — transcriptLanguage, isAutoGenerated and isTranslated on every row, so you know whether you got a person's captions, YouTube's speech recognition or a translation.
• The hook on every row — hook3s is the first 3 seconds of speech and hookStartSeconds says when the talking starts, so openers compare side by side without reading transcripts.
• Chain any YouTube scraper — pass a finished run's datasetId (or paste rows in datasetItems) and every video or channel link in those rows is transcribed; the row shape does not matter.
• Scheduled runs bill only what is new — name a watchlistId and switch on newItemsOnly; a video transcribed on an earlier run is left out, and a channel walk transcribes only its new uploads.
• Every miss on the record — asked = delivered + reported. A video that could not be transcribed is in the run's ERRORS record with a reason and what to change, not billed as a row, and the status line reconciles the count.

🎯 Use cases

Who	What they do with it
AI & LLM builders	Feed clean plain-text transcripts into RAG pipelines, summarizers, and agents (MCP-friendly output).
Content & SEO teams	Repurpose videos into articles, show notes, and quote pulls; mine competitor channels for topics.
Researchers & analysts	Build searchable corpora from talks, interviews and news coverage, with timestamps intact.
Subtitle & localization teams	Get SRT/VTT straight from the source, plus translated tracks where YouTube offers them.
Media monitoring	Track what's being said about brands and people across YouTube at scale.

📥 Supported inputs

Input	Example
Standard video URLs	https://www.youtube.com/watch?v=dQw4w9WgXcQ
Short links	https://youtu.be/dQw4w9WgXcQ
Shorts	https://www.youtube.com/shorts/{id}
Live stream replays	https://www.youtube.com/live/{id}
Channel	https://www.youtube.com/@handle (or /channel/…, /c/…, /user/…, its /videos or /streams tab) — walked newest first up to maxItems
Playlist	https://www.youtube.com/playlist?list=PL… — every playable entry, in playlist order, up to maxItems
Bare video id	dQw4w9WgXcQ
Chained dataset	datasetId: "nGkT…" — the default dataset of a finished YouTube scraper run; rows are deep-scanned for video and channel links
Pasted rows	datasetItems: [ { "url": "https://youtu.be/…" } ]

Not supported: private, members-only or age-gated videos. A stream still in progress has no captions yet; it is reported, not charged. A watch link that carries &list= is one video; use the playlist?list= link for the whole playlist.

🔄 How a run works

1. Each URL is resolved to its video ID; channels and playlists are expanded into their videos.
2. The caption track in your requested language is located (translated when that language is missing).
3. If that comes back empty, the run retries automatically by other routes, and with whisperFallback on, transcribes the audio as a last step.
4. Segments are normalised to timestamped JSON and rendered to SRT and VTT.
5. One row per video is pushed — transcript, formats, and metadata together.

⚙️ Input parameters

Field	Type	Default	Notes
startUrls	array	—	Video, Shorts, live, channel or playlist URLs, or bare video ids
videoUrl / urls	string / array	—	Accepted as well, so an input pasted from another transcript actor runs as is (not shown in the form)
datasetId	string	—	Default dataset of a finished YouTube scraper run; every video or channel link in its rows is transcribed
datasetItems	array	—	Rows pasted from a scraper instead of chaining by id
watchlistId	string	—	Named memory for scheduled runs; videos transcribed under this id are remembered
newItemsOnly	boolean	false	Leave out videos already on the watchlist. Needs watchlistId
language	string	default	Caption language code (en, de, …); default = the video's original track. targetLanguage is read too
whisperFallback	boolean	false	Whisper AI speech-to-text for caption-less videos — billed per transcribed video as a premium event
maxItems	integer	100	Hard cap on billed transcript rows
maxConcurrency	integer	10	Parallel video fetches
proxy	object	Automatic	Works out of the box; nothing to configure. Your own proxy URLs are used when you supply them

📊 Output overview

One row per transcribed video. The transcript appears three ways — transcript (timestamped segments), transcript_only_text (plain text), and transcript_srt / transcript_vtt (ready-to-save subtitle files) — alongside the video's metadata and the spoken hook. A video that could not be transcribed produces no row and no charge; it is listed in the run's ERRORS record with the reason (see below).

🧾 What was not delivered, and why

Every video the run set out to transcribe ends as a dataset row with a transcript or as an entry in the run's ERRORS record (key-value store), never neither, and never as a billed row without a transcript. The record is written on every run, with an empty items list when everything was delivered, so an integration can read it unconditionally. Each entry carries a reason, a plain-language detail with what to change, and a retryable flag:

reason	What happened	Re-run as-is?
no_captions	No caption track and whisperFallback is off	Switch on whisperFallback
no_speech	No captions, and Whisper heard only music, silence or noise	No
too_long	No captions, and the video is longer than Whisper's 30-minute cap	No
live_not_ready	A live stream still in progress, not started yet, or ended in the last day before YouTube published captions	Later: captions usually appear a few hours after the stream ends
unavailable	Private, removed, members-only or age-gated	No
login_required	YouTube asked the fetch to sign in	Yes
rate_limited	YouTube throttled the fetch	Yes, in a few minutes
fetch_failed	Another transport failure	Yes

The status line carries the same arithmetic, for example "Delivered 48 of 50 videos. 2 not delivered and not charged: 2 no captions and no Whisper fallback — see the ERRORS record." Videos the watchlist left out are counted separately and never charged. A run that delivers nothing because every fetch failed on YouTube's side is marked FAILED so it stands out in your run list.

🔁 Coming from pintostudio?

Paste your input as it is. Nothing to rename:

pintostudio/youtube-transcript-scraper	This actor
videoUrl (one video per run)	videoUrl works; startUrls takes a list of videos, channels and playlists in one run
targetLanguage (required, default en)	targetLanguage works; language is the same thing and defaults to the video's own track
Output data: [{ start, dur, text }]	The same data array on every row, plus plain text, SRT, VTT and the video's metadata
$0.01 per result	$0.005 per transcript, charged only when the row has one
Video without captions: error, empty data	Listed in ERRORS with a reason, not charged; opt-in Whisper transcribes it instead
Live replays fixed in October 2026	youtube.com/live/… links work, a stream still live is reported as live_not_ready
No channel or playlist input	Channels and playlists walked up to maxItems; a schedule with watchlistId bills only new uploads
No word on the track used	transcriptLanguage, isAutoGenerated, isTranslated, and every track in availableCaptionTracks

🔁 Coming from another YouTube transcript scraper?

What users of other transcript actors report on their issue boards, and what this one does about it:

Reported elsewhere	Here
Live video links fail with "impossible to retrieve video ID"	youtube.com/live/… links are read as-is; a stream still in progress is reported as live_not_ready, not billed
"No captions found" and an empty result	Opt-in Whisper transcribes caption-less videos up to 30 minutes; a video left untranscribed is named in ERRORS with a reason and never billed
Charged for empty results	A row is written, and billed, only when it carries a transcript
Want only a channel's new videos each day	watchlistId + newItemsOnly: a scheduled run transcribes and bills only uploads it has not seen
Want the publish date in the output	publishDate and uploadDate on every row
Want the whole transcript as one string, no timestamps	transcript_only_text
Want to see which caption tracks exist before choosing	availableCaptionTracks lists every language and whether it is auto-generated
Requested language ignored, always English	language returns the transcript in the language you ask for (checked with de and fr on an English video)

📦 Output sample

Real trimmed row from a live run:

{
    "videoId": "dQw4w9WgXcQ",
    "title": "Rick Astley - Never Gonna Give You Up (Official Video) (4K Remaster)",
    "author": "Rick Astley",
    "channelId": "UCuAXFkgsw1L7xaCfnd5JJOw",
    "lengthSeconds": "213",
    "viewCount": "1699540216",
    "transcript": [
        { "text": "[♪♪♪]", "startMs": "1360", "endMs": "3040", "startTimeText": "0:01" },
        { "text": "♪ We're no strangers to love ♪", "startMs": "18800", "endMs": "22140", "startTimeText": "0:18" }
    ],
    "transcript_only_text": "[♪♪♪] ♪ We're no strangers to love ♪ ♪ You know the rules and so do I ♪ …",
    "transcript_srt": "1\n00:00:01,360 --> 00:00:03,040\n[♪♪♪]\n…",
    "transcript_vtt": "WEBVTT\n\n00:00:01.360 --> 00:00:03.040\n[♪♪♪]\n…",
    "data": [{ "start": "1.36", "dur": "1.68", "text": "[♪♪♪]" }, { "start": "18.64", "dur": "3.24", "text": "♪ We're no strangers to love ♪" }],
    "transcriptSource": "timedtext",
    "transcriptLanguage": "en",
    "isAutoGenerated": false,
    "isTranslated": false,
    "keywords": ["rick astley", "Never Gonna Give You Up", "nggyu"],
    "thumbnail": { "thumbnails": [{ "url": "https://i.ytimg.com/vi/dQw4w9WgXcQ/…", "width": 168, "height": 94 }] }
}

🗂 Key output fields

Field	Meaning
transcript[]	Timestamped segments: text, startMs, endMs, startTimeText
transcript_only_text	The whole transcript as one plain string — LLM-ready
transcript_srt / transcript_vtt	Complete subtitle files as strings — save and use directly
transcriptSource	Which step produced it: one of the caption routes, or whisper when the audio was transcribed
transcriptLanguage	Language code of the text you got; null when the source did not say
isAutoGenerated	true for YouTube's speech recognition, false for a person's captions, null when it cannot be told
isTranslated	true when the text was translated into the language you asked for
data[]	The same segments as start, dur, text in seconds (as strings), the shape pintostudio's actor returns
availableCaptionTracks[]	Every caption track on the video: languageCode, name, kind (manual or auto-generated), isTranslatable
publishDate / uploadDate	When the video was published
channelHandle	The channel's @handle, from its profile link
liveBroadcastDetails	Start and end time for a video that was a live stream
hook3s / hookStartSeconds	The first 3 seconds of speech, and when speech starts; a music-only opening reads as the caption marker (for example [♪♪♪])
videoId, title, author, channelId	Video identity
viewCount, lengthSeconds, keywords, shortDescription, thumbnail	Metadata that rides along free

❓ FAQ

What happens with videos that have no captions? Without whisperFallback they produce no row and no charge, and the ERRORS record lists them as no_captions. With whisperFallback: true, the audio is downloaded and transcribed by Whisper speech-to-text — billed as a separate premium event per video, only when it actually produces a transcript. Over music or silence Whisper tends to invent filler ("Thank you.", "♪"); a result made only of that counts as no transcript, so the video is listed as no_speech and neither the row nor Whisper is charged. Whisper handles videos up to 30 minutes long.

Which languages are supported? Any language YouTube has a caption track for. Set language to a code like de or es; when that exact track is missing, YouTube's auto-translated track is used where available.

Can I transcribe a whole channel or playlist? Yes. A channel URL (youtube.com/@handle) is walked newest first and a playlist URL (youtube.com/playlist?list=…) in playlist order, both up to maxItems. For search results, run a YouTube search scraper and pass its dataset id in datasetId; every video link in the rows is transcribed. On a schedule, add a watchlistId with newItemsOnly so only new uploads are transcribed and billed.

My integration starts runs with a short timeout. What happens to a long run? About 45 seconds before the run's timeout the actor stops taking new videos, finishes the ones in progress and ends normally. Rows already delivered stay in the dataset, and the status line says the run stopped early. Unfinished videos are not charged.

Do I need to configure proxies? No. The default works out of the box at any volume. If you want runs to go through your own proxies, switch off "Use Apify Proxy" in the proxy input and add your proxy URLs.

Do failed videos cost me anything? No. A video without a transcript produces no row, so nothing is billed for it; it is named in the ERRORS record instead. Whisper is only charged when it delivers text.

💵 Pricing

Event	When	Price
Transcript	A row with a transcript is written	$0.005
Whisper AI transcription	Opt-in, only when Whisper produces text for a caption-less video	$0.20
Actor start	Once per run	$0.00005

A video that comes back without a transcript writes no row and costs nothing.

💬 Support

Found a bug or missing a field? Open an issue on the actor's Issues tab in Apify Console — issues are answered within 1–2 business days.

🛠 Additional services

Need scheduled transcript archives, a merged multi-platform transcript feed (YouTube + TikTok + Instagram + Loom), or delivery straight to your database? Custom builds and SLAs available — contact me through the actor page.

🔎 Explore more scrapers

Same developer, same stack: Video & Audio Transcriber (Whisper), Instagram Transcript Scraper, YouTube Comments, YouTube Search — and the full portfolio at memo23 on Apify Store.

🤖 For AI Agents & LLM Apps

Minimal input: { "startUrls": ["https://www.youtube.com/watch?v=dQw4w9WgXcQ"] } (or { "videoUrl": "…" }). One row per video; read transcript_only_text for the text and the run's ERRORS record for anything not delivered. Built for machine consumption: transcript_only_text drops straight into a context window; timestamped segments support citation and chaptering; stable field names across every row. Pair with the Video Transcripts MCP Server to expose transcripts as a tool in agent frameworks. Keep maxItems low per call for cost control; every row is self-contained.

---

⚠️ Disclaimer

This Actor is an independent tool and is not affiliated with, endorsed by, or sponsored by YouTube or Google LLC. All trademarks mentioned are the property of their respective owners.

The scraper accesses only publicly available video pages and caption data — no login, no age-gated, members-only or private content. Users are responsible for ensuring their use complies with YouTube's Terms of Service, copyright law as it applies to transcript content, applicable data-protection law (GDPR, CCPA, etc.), and any contractual obligations of their own organisation.

---

SEO Keywords

youtube transcript scraper, youtube transcript api, extract youtube transcript, youtube captions scraper, youtube subtitles downloader, srt from youtube, vtt from youtube, youtube video to text, youtube transcription tool, whisper youtube transcription, transcribe youtube videos without captions, youtube transcript for llm, youtube rag pipeline, bulk youtube transcripts, youtube caption extractor, video to text api, youtube shorts transcript, youtube transcript json, youtube playlist transcript, youtube channel transcripts, auto-generated captions, apify youtube transcript, pintostudio alternative, youtube-transcript-scraper alternative