TikTok, Instagram & YouTube Transcript — real Whisper ASR

Transcribes the actual audio with Whisper large-v3. Not captions.

Most transcript actors read the subtitles a platform already published. That works on YouTube videos that happen to have captions — and returns nothing otherwise. This one downloads the audio and runs speech recognition on it, so it behaves the same whether captions exist or not.

What it handles that caption-readers don't

• Videos with no subtitles at all — the majority of TikTok and Instagram.
• TikTok and Instagram, not just YouTube.
• Any language, auto-detected, including videos whose captions are only available in one language.
• Auto-generated captions you don't trust — this reads the audio directly.

Pricing

$0.004 per minute of audio transcribed. Rounded up to the minute.

That is roughly 6× cheaper than other actors that run real speech recognition (they charge $0.025–$0.05 per minute), because transcription runs on dedicated GPUs instead of inside the Actor.

You are not charged when there is nothing to transcribe — a silent video, a photo post, or a carousel returns a labelled row at no cost.

Input

{
  "urls": [
    "https://www.tiktok.com/@nasa/video/7670721000471891214",
    "https://www.instagram.com/nasa/reel/Dbn-XJhk0_-/",
    "https://www.youtube.com/watch?v=aircAruvnKk"
  ],
  "language": null,
  "includeSegments": true,
  "maxDurationSeconds": 3600
}

Field	Meaning
urls	Video URLs. YouTube, TikTok, Instagram.
language	ISO code (en, fr, es…). Leave empty to auto-detect — detection is reliable and free.
includeSegments	Sentence-level segments with start/end times.
maxDurationSeconds	Skip anything longer. Guards against transcribing a multi-hour stream by accident.
instagramCookies	Instagram only. See below.

Instagram requires a session

Instagram returns an empty media response to anonymous requests — every public post, no exceptions. Export cookies from a logged-in browser (Netscape format) and pass them in instagramCookies. They are written to a temporary file and deleted when the run ends; they are never stored.

YouTube and TikTok need no credentials.

Output

{
  "url": "https://www.tiktok.com/@nasa/video/7670721000471891214",
  "platform": "tiktok",
  "videoId": "7670721000471891214",
  "title": "There's nothing like watching humanity leave Earth.",
  "uploader": "nasa",
  "durationSeconds": 29.93,
  "language": "en",
  "languageConfidence": 0.962,
  "transcript": "Booster ignition and lift off, the crew of Artemis 2 now bound for the moon…",
  "segments": [{ "debut": 0.0, "fin": 4.2, "texte": "Booster ignition and lift off" }],
  "source": "audio-asr",
  "scrapedAt": "2026-08-06T04:10:18.295Z"
}

source is always audio-asr — it is there so you can verify the text came from speech recognition rather than from a caption file.

Rows that could not be transcribed carry errorType:

errorType	Meaning
no_audio_track	Silent video, or the post is a photo / carousel. Not charged.
extraction_failed	The video could not be downloaded — private, deleted, or region-locked.

Notes

Long videos are fine: a 19-minute YouTube video returns 257 timestamped segments. Very long recordings are limited by maxDurationSeconds, which you can raise.

Residential proxies are recommended for TikTok and Instagram, which rate-limit datacenter IP ranges.