Audio & Video Transcriber (Whisper, SRT, TikTok, Podcasts)

Price: about $0.04 per hour of audio ($0.0006 per minute, billed per second), plus a tiny per-run start fee. Speaker labels add the same again.

Turn an audio or video file link, or the page link of a TikTok, Instagram or X video, into text, in about 99 languages (auto-detected, including Mandarin). Optional extras:

• timestamped segments with confidence;
• word-level timings;
• speaker labels (beta: who spoke when);
• SRT and WebVTT subtitles.

Two kinds of link work:

• A file link (.mp4, .m4a, .mp3, .webm, .ogg ...): the most reliable, for example the download URL a TikTok, Instagram, Douyin or podcast scraper returns.
• A page link such as https://www.tiktok.com/@user/video/123...: the audio is downloaded for you inside the run and removed again right after transcription. This is best effort: some platforms refuse cloud servers (a few TikTok posts, private or login-only posts and YouTube links currently fail). When a link can't be downloaded the run stops with a clear message and only Apify's start fee is charged.

What you get

One dataset item per run:

{
  "text": "NASA's Curious Universe. Houston, we have a podcast. ...",
  "language": "en",
  "language_probability": 0.996,
  "duration_seconds": 62.84,
  "processed_seconds": 62.84,
  "truncated": false,
  "complete": true,
  "segments": [{"start": 0.0, "end": 1.48, "text": "NASA's Curious Universe.", "confidence": 0.93, "speaker": "S1"}],
  "words": [{"start": 0.0, "end": 0.54, "word": "NASA's", "probability": 0.994, "speaker": "S1"}],
  "speakers": 2,
  "turns": [{"speaker": "S1", "start": 0.0, "end": 14.6, "text": "NASA's Curious Universe. ..."}],
  "srt": "1\n00:00:00,000 --> 00:00:01,480\nS1: NASA's Curious Universe.\n...",
  "billed_seconds": 63
}

Input

Field	Meaning
url	Direct http(s) link to an audio or video file (required)
max_minutes	Transcribe and pay for at most this many minutes (1–30). Longer media is cut and flagged truncated.
language	Whisper code (en, es, zh, ...); leave empty to auto-detect
segments, words, srt, vtt	Extra outputs
speakers	Speaker labels (beta, paid add-on, up to 20 minutes per run)
num_speakers	The number of speakers, if known; improves speaker labels

Pricing

Pay per event, only for audio actually transcribed:

• audio-second: per second of processed audio;
• speaker-second: the speaker-label add-on, per second, charged only when labels are returned.

A failed run (for example a web page instead of a media file) charges no audio or speaker events, only Apify's standard run-start fee ($0.00005).

Accuracy and speed

• Engine: Whisper large-v3-turbo.
• Accuracy: about 98% of words correct on clear speech; about 1–2% word error rate on the JFK inaugural address.
• Speed: a 1-minute clip takes seconds, and 30 minutes takes up to ~4 minutes at peak.
• Speaker labels are best-effort. They work well for clearly different voices; pass num_speakers when you know it.

Privacy

The media is deleted as soon as its audio is decoded, and transcripts are never stored or logged by the transcription service. For page links, the downloaded audio is held in your run's storage only while it is transcribed, then deleted. The result lives only in your run's dataset.