Media Transcriber — audio & video to text, SRT and VTT

Give it a link to an audio or video file. Get back the transcript, timestamped segments, and ready-to-use subtitle files.

No API key. No ffmpeg install. No GPU. The speech model runs inside the Actor, so there is nothing to set up and no second bill from an external provider.

What you get

For every file you submit:

Field	What it is
text	The full transcript as one string
segments	[{start, end, text}] — seconds, ready to seek with
language	Detected language code, plus languageProbability
durationSeconds	Length of the media
srtKey / vttKey	Key-value store keys for the subtitle files

SRT and VTT are written to the run's key-value store as downloadable files.

Input

{
  "mediaUrls": [
    "https://example.com/interview.mp3",
    "https://example.com/webinar.mp4"
  ],
  "model": "base",
  "language": "",
  "translateToEnglish": false,
  "outputFormats": ["srt", "vtt"],
  "maxFileSizeMb": 500
}

• mediaUrls — upload files directly, or paste links to them. MP3, MP4, WAV, M4A, OGG, FLAC, WebM, MOV and anything else FFmpeg can open. A link must point straight at the file.
• model — tiny, base, small, or medium. Start with base. Move up if the audio has accents, background noise, or is not in English.
• language — ISO-639-1 code such as en, ja, es. Leave empty to auto-detect. Setting it explicitly is faster and more reliable on short clips.
• translateToEnglish — translate the speech into English instead of transcribing it as spoken.
• maxFileSizeMb — hard cap, so a mistyped link cannot run up a bill.

Pricing

Pay per event, billed per minute of audio — not per run, not per file.

Event	Price
Run start	$0.005
Minute of audio, tiny / base	$0.02
Minute of audio, small / medium	$0.05

Audio is rounded up to the next whole minute per file. A 3½-minute podcast on base costs $0.005 + 4 × $0.02 = $0.085.

Use it from an AI agent (MCP)

This Actor is exposed as an MCP tool, so an agent can transcribe media without you writing any glue code. Point your MCP client at Apify's server and call this Actor by name with mediaUrls.

What it will not do

Links to video-sharing and social platforms are rejected. YouTube, TikTok, Instagram, Facebook, X, Vimeo, SoundCloud, Spotify, Twitch and similar sites forbid automated downloading in their terms of service, so this Actor refuses those hosts rather than leaving you exposed.

A paid subscription does not change this. Offline downloads are licensed for playback inside that platform's own app, not for extraction and reuse.

Use a file you hold the rights to: your own recordings, your podcast's CDN, a client's upload bucket, a public-domain archive — or just upload the file.

URLs that resolve to private or link-local addresses are rejected too, so the Actor cannot be pointed at an internal network.

Notes

• Silence and music return zero segments. That is voice-activity detection working correctly, not a failure.
• Very long files are fine; cost scales linearly with duration.
• Files that fail are reported in the dataset with status and error, and the run continues with the remaining URLs.

---

Built with faster-whisper. Developed with Claude Code; behaviour verified against real audio by the author.