Podcast transcription from the episode URL: parts and cost

Podcast transcription turns an episode's audio into text for show notes, search, and accessibility. A published episode already sits at a public audio URL, the enclosure link in the show's RSS feed, so a speech to text API that takes URLs can read it from there with no download and re-upload. Sume's STT 1.0 does that for up to 10 minutes of audio per request: a longer episode goes in parts, and the transcript has word timestamps but no speaker labels.

The facts come from the STT 1.0 schema in the Sume API reference, the OpenAPI document behind the API reference docs, and the price from the API rate card, read on 2026-09-29.

Can I transcribe a podcast straight from its URL?

Yes, if the episode is 10 minutes or shorter and its URL is public HTTPS. Put the enclosure URL in audio_url and submit; the job reads the file itself. If the enclosure is a redirecting link, the docs don't say whether redirects are followed, so test one episode first or pass the final file URL.

curl -X POST https://api.sume.com/v1/stt-1.0/transcribe \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: episode-118" \
  -d '{
    "audio_url": "https://example.com/podcast/episode-118.mp3",
    "duration_seconds": 540,
    "language_code": "en",
    "segmentation": { "mode": "sentence" }
  }'

How do I transcribe a full-length episode?

Cut it into parts of 10 minutes or less, put each part at a public HTTPS URL in your own storage, and submit one request per part. Sume can't cut a remote episode for you: its Timeline audio split works only on Sume-hosted media.sume.com audio, and there is no public route for uploading local files. Each part's timestamps are seconds from the start of that part, so add the part's start time before you merge them. Transcribe long audio files walks through the split and the offset math.

For a back catalog, submit every part at once with its own Idempotency-Key and a webhook_url for the terminal callback, keeping status_url polling as a backup. Concurrency is a dispatch limit, not a submit limit: parts above your workspace's concurrency limit are accepted as queued while queue capacity remains.

What does a podcast transcript from the API include?

• text: the whole transcript of the part.
• words[]: each word, with start and end in seconds when they are returned (current code), useful for search or clip timecodes.
• segments[], when you send segmentation with mode sentence: gapless, ordered sentences with start and end times, a base for show-notes paragraphs or chapters.
• No speaker labels: diarize is fixed server-side. For host and guest names, see how to transcribe an interview.
• The language is auto-detected unless you pass language_code.

How much does podcast transcription cost?

STT 1.0 costs $0.01 per audio minute, plus a 5.5% agent fee by default. Send each part's duration_seconds: without it, a request reserves 1 minute. The rows below apply that rate to three episode lengths, before the fee.