Audio & Video Transcriber: Diarization, Subtitles & RAG

A paid Apify Actor for batch transcription of direct audio/video URLs and uploaded files. Actor users do not need to bring their own transcription-provider API key.

The first release intentionally focuses on a reliable media-to-transcript core. Social-page extraction is kept behind a future source-adapter boundary so a YouTube or TikTok layout change cannot break ordinary file transcription.

Why this Actor is worth paying for

• One run accepts up to 100 direct URLs and uploads, with bounded concurrency and per-file failure isolation.
• standard mode handles ordinary recordings; conversation adds speaker diarization.
• Names and specialist vocabulary can be boosted with a custom glossary.
• Output includes confidence, review ranges, speakers, timestamps, detected language, and glossary coverage.
• Each successful item can produce TXT, Markdown, normalized JSON, SRT, and WebVTT files.
• Optional RAG rows preserve source URL, start/end timestamps, speaker IDs, token estimate, and a stable content hash.
• Provider-side PII/PCI/PHI/number redaction and Model Improvement Program opt-out are exposed explicitly.
• Cost guards reject oversized or over-duration media before transcription. Pay-per-event charges happen only after requested artifacts and RAG rows are written.
• Signed URL query strings are removed from logs and output URLs. Direct URLs are restricted to public HTTP(S) hosts to reduce SSRF risk.

Input example

{
  "mediaUrls": [
    "https://example.com/recording.mp3"
  ],
  "mode": "conversation",
  "language": "auto",
  "glossary": ["Apify", "your product name"],
  "redact": [],
  "privacyMode": true,
  "outputFormats": ["txt", "md", "json", "srt", "vtt"],
  "createRagChunks": true,
  "maxMinutesPerItem": 240,
  "maxTotalMinutes": 600
}

Set dryRun to true to download and inspect the files, return durations and estimated billed minutes, and make no transcription request or transcription-event charge.

language accepts auto, multi for code-switching, or a supported language code. Entity redaction (pii, pci, phi) is currently English-only; number redaction supports additional languages.

Output contract

The default Dataset gets one result row per input:

• success: complete normalized transcript, compact segments, metadata, artifact links, RAG link, and billing details.
• validated: dry-run media metadata and estimated minutes.
• failed: stable error code, stage, retryability, and zero billed minutes.
• duplicate: duplicate input/options combination skipped with zero billed minutes.

Artifacts are stored in the default Key-Value Store. RAG chunks are stored in audio-transcriber-rag-<run ID> when enabled. The full raw provider response is retained only when storeRawProviderResponse is true.

Pay-per-event setup

Create these events in the Actor's Apify Console monetization settings before publishing:

Event	Suggested launch price	Charged when
apify-actor-start	$0.00005/run	Apify's synthetic Actor-start event; keep the recommended default
standard-minute	$0.015/min	A standard transcript and all requested outputs succeeded
conversation-minute	$0.022/min	A diarized transcript and all requested outputs succeeded
redacted-minute	$0.004/min	Successful output requested one or more redaction groups

In Console, open Development → My Actors → this Actor → Publication → Monetization → Set up monetization, choose pay per event, and add the three custom minute events exactly as written. Remove the synthetic apify-default-dataset-item event so the result row is not billed in addition to the transcription event. Apify charges apify-actor-start automatically; the code must not charge it.

Durations are rounded up separately per file. Glossary processing cost is absorbed by the transcription event rather than exposed as a confusing fourth usage add-on. Revisit prices after measuring the real mix of multilingual audio, privacy mode, diarization, retries, and Apify compute.

The Actor charges only for completed items and preflights Apify's maximum-run-charge limit before sending media to the transcription provider. Configure all three code-charged event names exactly as shown; a missing event is a deployment configuration error.

Release boundary

Version 0.1 supports direct, publicly reachable media URLs and Apify file uploads. It does not claim to extract media from arbitrary webpage, YouTube, TikTok, Instagram, Google Drive share, or Dropbox share URLs. Those should be introduced as individually tested source adapters, with platform-specific terms and failure modes documented before they are marketed.

Privacy notes

• Files are downloaded into an isolated temporary directory and deleted when processing finishes.
• Signed URL queries are used for download but removed from result rows, logs, Markdown, and RAG provenance.
• privacyMode=true requests that the transcription provider exclude media from model-improvement programs where supported.
• Redaction happens at the provider before the transcript is returned, but media still leaves Apify for third-party transcription processing.
• Apify storage retention and access remain controlled by the Actor run owner.

This is not a compliance certification. Customers handling regulated data should review Apify's and their own compliance requirements before use.