Audio & Video Transcriber — any file or link, no API key

Give it a URL to a sound, get the words back. A podcast episode, a file sitting on S3 or your own server, a recorded interview, a Vimeo video, a lecture — anything you can link to.

Speech recognition is included in the price. No OpenAI key, no Deepgram key, no second bill to reconcile.

in:   https://example.com/episode-142.mp3
out:  full text · timed segments · a ready-to-use .vtt subtitle file

Why "no API key" is the whole point

Most transcription actors on the store are BYOK — bring your own key. To use one you sign up with an AI provider, put a card on file, generate a key, paste it into the actor input, and then watch two bills instead of one. If you are evaluating three tools, that is three signups before you learn whether any of them works.

Here you paste a URL.

What it handles

input	what happens
direct .mp3, .wav, .m4a, .mp4	transcribed. A direct file has no metadata page, so the duration is measured from the file rather than refused
Vimeo, YouTube, Twitch, SoundCloud (~1800 sites)	if real subtitles already exist they are read for free; otherwise the audio is transcribed
a two-hour recording	cut into overlapping windows, transcribed, reassembled onto the original clock — timings stay correct at the end
a playlist or channel URL	refused, with the reason. It could be thousands of files and you would be billed for all of them

What you get, per file

{
  "url": "https://example.com/episode-142.mp3",
  "durationSeconds": 3612,
  "source": "whisper",
  "isAiGenerated": true,
  "language": "en",
  "text": "the full transcript…",
  "wordCount": 8213,
  "segments": [{ "start": 0.2, "end": 3.1, "text": "…" }],
  "vtt": "WEBVTT…",
  "lowConfidence": false,
  "warning": null,
  "hallucinationsRemoved": 0
}

Three things it refuses to do, on purpose

1. It never returns a partial transcript. Long audio is cut into windows and reassembled, and if any window goes missing the file fails with an error rather than returning text with a silent hole in the middle. A transcript with a gap reads perfectly and is undetectable by you — which is exactly what makes it the one output worth failing to avoid.

2. It never bills you for work it did not do. If a page already has real subtitles, they are read and no transcription minutes are charged. A failed file writes a row explaining which URL failed and why, and is charged nothing at all — one dead link in a batch of 200 never costs you the other 199.

3. It tells you when it could not really hear anything. Speech models trained on subtitle files invent subtitle furniture over silence and music — translator credits, "thanks for watching". Those are stripped and counted in hallucinationsRemoved, and if little real speech is left the row comes back lowConfidence: true with a reason. An honest "we could not hear speech in this" is more useful than a fluent invention, and you can filter on it.

Input

{
  "videoUrls": ["https://download.samplelib.com/mp3/sample-15s.mp3"],
  "language": "en",
  "includeSubtitles": true,
  "includeCutList": false
}

includeCutList returns the spans of dead air, for editors cutting long footage.

Pricing

Per result, plus per minute actually transcribed — so a short clip costs like a short clip, and you are never charged transcription minutes for subtitles that already existed.