Podcast Transcription API — RSS feeds to episode transcripts

Point it at a podcast RSS feed and get transcripts of the episodes you actually want — the newest N of each show, everything published after a date, or only episodes whose title matches a phrase. Whisper runs inside the Actor (faster-whisper, int8, CPU), so there is no API key, no external service, and no per-provider markup. 99 languages, auto-detected. Optional translation to English.

Every item carries the show and episode metadata from the feed — show title, author, episode number, season, publish date, GUID — next to the transcript, so a podcast pipeline never has to resolve enclosure URLs or re-fetch the feed to find out what it just transcribed.

Built for podcast search, show-notes generation, clip finding, media monitoring and AI agents that need episodes as text.

What you get

One dataset item per episode:

{
  "feed_title": "Gravity Assist",
  "feed_author": "Brian Dunbar",
  "episode_title": "Season 5, Episode 32: Finale",
  "episode_number": 32,
  "season": 5,
  "published": "2022-08-12T18:21:00+00:00",
  "episode_guid": "…",
  "url": "https://www.nasa.gov/…/gafinale081222.mp3",
  "status": "ok",
  "language": "en",
  "duration_seconds": 1864.3,
  "billed_minutes": 32,
  "n_segments": 214,
  "n_words": 4930,
  "text": "Full transcript as one string …",
  "segments": [
    { "start": 0.0, "end": 6.48, "text": "Welcome back to the show." }
  ],
  "srt": "1\n00:00:00,000 --> 00:00:06,480\nWelcome back to the show.\n",
  "vtt": "WEBVTT\n\n00:00:00.000 --> 00:00:06.480\nWelcome back to the show.\n"
}

Episode selection

Field	Default	What it does
feed_urls	—	up to 10 RSS/Atom feeds, 40 MB each
episodes_per_feed	3	how many matching episodes to take per show
published_after	—	2026-08-01, or an ISO timestamp
title_contains	—	case-insensitive substring match on the title
episode_urls	—	extra direct audio URLs, transcribed alongside the feeds

Filters combine: published_after and title_contains must both match. Selection runs in feed order — which is newest-first for virtually every show — and then takes the first episodes_per_feed.

Two deliberate behaviours worth knowing:

• An episode whose feed gives no usable date survives a published_after filter. Dropping it would hide episodes because the publisher's XML is sloppy, which is not your problem to debug.
• A direct episode URL pasted into feed_urls is detected and transcribed as a single episode instead of failing as a malformed feed.

Up to 50 episodes are queued per run, however the feeds divide it up.

Pricing — pay only for delivered transcripts

Event	Price	Charged when
episode-transcribed	$0.005	an episode delivered a non-empty transcript
transcription-minute	$0.008	per started minute of audio transcribed

A 30-minute episode costs $0.245. A weekly show, 4 episodes a month, runs about $1.

There is no actor-start fee — a start fee bills you for failed runs, and this family never charges one.

The billing rule is deliberate: episodes that fail to download, cannot be decoded, or contain no recognizable speech are never billed — not even the per-episode fee. A feed that will not parse is reported in the dataset and costs nothing. Your cost ceiling per episode = $0.005 + $0.008 × Max minutes per episode (default 120).

Transcription options

Field	Default	Notes
model	base	tiny / base / small — accuracy vs speed
language	auto	ISO code (en, ar, es, …) skips detection
task	transcribe	translate = English text from any language
vad_filter	true	skip silence — faster, fewer hallucinated lines
max_minutes_per_episode	120	hard cost/length cap; longer episodes flagged truncated

tiny and base are baked into the image and start instantly; small downloads once at the start of the run (~1 minute extra).

Re-running without paying twice

Each run checkpoints the episode URLs it finished, so a migrated or resumed run never re-transcribes — and never re-charges — work it already delivered. Across runs, deduplicate on episode_guid: it is the identifier the feed itself considers stable.

Limits, stated plainly

• CPU transcription: roughly 5–15 seconds of processing per audio minute with base, more with small. A 60-minute run timeout comfortably covers three 45-minute episodes with base; raise the timeout for longer batches or for small.
• 10 feeds per run, 25 episodes per feed, 50 episodes per run, 250 MB per episode, 480 minutes max per episode.
• Video podcasts work — the audio track is what gets transcribed.
• Feeds behind a paywall or private RSS token work only if the enclosure URL itself is fetchable without a login.

The family

Actor	Use it for
Podcast Transcription API (this one)	RSS feeds → episode transcripts + metadata
Whisper Transcriber	any audio/video file → transcript text
SRT Subtitles Generator	media → finished .srt / .vtt caption files
PDF OCR Extractor	scanned PDFs, Arabic + English OCR

Issues and requests

Open an issue on the actor's Issues tab — I read all of them.