Podcast Transcript Scraper 🎙️ With Speaker Labels

Transcripts that podcast publishers provide themselves, in the open Podcasting 2.0 standard. Millisecond timestamps and speaker labels — the two things video captions almost never have.

What it does

• Download podcast transcripts in bulk
• Get timestamped segments with speaker labels
• Build a searchable archive of a show's back catalogue
• Export podcast transcripts to JSON or CSV
• Feed spoken-word text into a RAG pipeline or a language model

Why these are better than video captions

Publishers ship these deliberately, produced properly rather than guessed from audio:

00:00:00,240 --> 00:00:02,880
Speaker 1: Please note that this show discusses...

Two things come free that automatic video captions do not give you:

• Speaker labels. Diarisation is expensive to compute and here it is already done. hasSpeakerLabels and speakerCount are on every row.
• Real punctuation. These are edited files, not a stream of words.

The honest number

Roughly half of podcast feeds publish a transcript. Measured across 30 feeds sampled from eight genres on 4 September 2026, 14 of 30 — 46.7% — carried the <podcast:transcript> tag, at 114 to 171 tagged episodes per show.

The other half publish none, and no tool can extract text that was never created. Those episodes come back as no-transcript-published and are never charged. Anything claiming complete coverage of all podcasts is either running speech recognition — a different and far more expensive product — or is wrong.

Sample size is 30, so treat 46.7% as a first reading rather than a settled figure.

You only pay for transcripts

errorCode	Meaning	Charged
— (success: true)	A transcript	Yes
no-transcript-published	Publisher provides none	No
transcript-unreachable	Link published but the file would not fetch	No
transcript-empty	File fetched but had no readable cues	No
feed-unreachable	Feed did not respond or is not valid RSS	No
show-not-found	No podcast matched that name	No
no-episodes	Feed is valid but empty	No

Input

Field	Required	Description
shows	yes	Podcast names or RSS feed URLs
maxEpisodesPerShow	no	Newest first. Default 50. Your cost ceiling
language	no	Two-letter code. A filter, not a preference
includeSegments	no	Adds timed cues. Default false
maxConcurrency	no	1–15, default 5

{
  "shows": ["This Podcast Will Kill You", "https://feeds.megaphone.fm/vergecast"],
  "maxEpisodesPerShow": 100,
  "includeSegments": true
}

A feed URL skips the directory lookup, so it is faster than a name.

Output

Every row has the same fields whether it succeeded or failed, so you can select columns without branching.

Field	Type	Description
success	boolean	True when the row carries a transcript
sourceShow	string	The name or feed you supplied
showTitle	string	The show's own title
episodeTitle	string	Episode title
episodeGuid	string	Feed's unique id — deduplicates across runs
episodeUrl	string	Episode page, where published
publishedAt	string	Publication date
seasonNumber	integer	Where published
episodeNumber	integer	Where published
durationSeconds	number	From the feed, or the transcript's own span
transcriptFormat	string	MIME type the publisher provided
language	string	As the publisher declared it
hasSpeakerLabels	boolean	True when speakers are named
speakerCount	integer	Distinct speakers labelled
segmentCount	integer	Timed cues — always reported
fullText	string	Whole transcript as plain text
segments	array	Timed cues with speakers; null unless requested
errorCode	string	Null on success
errorMessage	string	Human-readable explanation

{
  "success": true,
  "sourceShow": "This Podcast Will Kill You",
  "episodeTitle": "Ep 141: Scurvy",
  "durationSeconds": 3723,
  "transcriptFormat": "application/srt",
  "hasSpeakerLabels": true,
  "speakerCount": 3,
  "segmentCount": 842,
  "fullText": "Hi. My name's Elan. I'm a nurse practitioner...",
  "errorCode": null,
  "errorMessage": null
}

Use it from an AI agent (MCP)

This Actor is callable as a tool through the Apify MCP server, so Claude, ChatGPT, Cursor and VS Code can run it directly.

Add the server to your MCP client:

{
  "mcpServers": {
    "apify": {
      "url": "https://mcp.apify.com",
      "headers": {
        "Authorization": "Bearer <YOUR_APIFY_TOKEN>"
      }
    }
  }
}

Then ask for what you want in plain language — for example “get the transcripts for this podcast” — and the agent calls cleanfeed/podcast-transcript-downloader with the right input. Every output field is described in the dataset schema, so the agent knows what it is getting back before it runs anything.

Call it from code

Python

from apify_client import ApifyClient

client = ApifyClient("<YOUR_APIFY_TOKEN>")

run = client.actor("cleanfeed/podcast-transcript-downloader").call(run_input={
    "shows": ["This Podcast Will Kill You"],
})

for item in client.dataset(run["defaultDatasetId"]).iterate_items():
    if item["success"]:
        print(item)

JavaScript

import { ApifyClient } from 'apify-client';

const client = new ApifyClient({ token: '<YOUR_APIFY_TOKEN>' });

const run = await client.actor('cleanfeed/podcast-transcript-downloader').call({
    shows: ["This Podcast Will Kill You"],
});

const { items } = await client.dataset(run.defaultDatasetId).listItems();
console.log(items.filter((i) => i.success));

cURL

curl -X POST "https://api.apify.com/v2/acts/cleanfeed~podcast-transcript-downloader/run-sync-get-dataset-items?token=<YOUR_APIFY_TOKEN>" \
  -H 'Content-Type: application/json' \
  -d '{"shows": ["This Podcast Will Kill You"]}'

Related actors

If you need	Use
Transcripts from YouTube videos instead	YouTube Transcript Scraper
Transcripts for a whole YouTube channel	YouTube Channel Transcript Scraper

Limitations

• Roughly half of feeds publish no transcript. Measured across 30 feeds in eight genres on 4 September 2026, 14 of 30 carried the <podcast:transcript> tag. The rest return errorCode: no-transcript-published and are never charged. Sample size is 30, so treat 46.7% as a first reading.
• No speech-to-text fallback. Transcripts are read, not generated. Recognition costs real money per hour of audio and would change the price by a large multiple.
• Transcripts are the publisher's work and remain theirs. Quote with attribution; do not republish an episode's transcript as your own content.
• Quality varies with who produced it. Some are professionally edited, some are lightly-corrected machine output. The format and speaker count are reported so you can judge; the text is never rewritten.
• Speaker labels are as the publisher wrote them, often "Speaker 1" rather than a real name, with no published mapping.
• Transcript links point at the publisher's host, not ours. A dead link is reported as transcript-unreachable rather than hidden.

FAQ

Why does this show return nothing?

Most likely the publisher does not include <podcast:transcript> in its feed. Around half do not. You get one row per episode saying so, at no charge.

Can you transcribe the audio when there is no transcript?

Not in this Actor. Speech recognition costs real money per hour of audio and would change the price by a large multiple. Shipping it silently inside a per-transcript price would be dishonest about what you are paying for.

What is the difference from a YouTube transcript scraper?

The source. YouTube gives automatic captions, which have no speaker labels and unreliable punctuation. These are files the publisher produced, usually with both.

Does it work for any podcast?

Any podcast with a public RSS feed, which is nearly all of them. Shows exclusive to one platform with no public feed cannot be reached.

Do I get speaker names?

You get the labels the publisher wrote. Many use Speaker 1 and Speaker 2 rather than real names, and no mapping is published. hasSpeakerLabels and speakerCount tell you what you have before you read the text.

Can I pass an RSS feed directly?

Yes, and it is faster — a feed URL skips the podcast-directory lookup entirely.

Notes

Only transcripts that publishers link from their own RSS feeds are downloaded, via the open Podcasting 2.0 <podcast:transcript> tag. No audio is processed and no transcript is generated — if a publisher provides none, that is reported.