"Podcast transcript API" means three different jobs, and most comparison lists mix them. Work out which one you're doing first; it eliminates most of this list.
- Retrieve a transcript that already exists: building an archive, a RAG corpus, a summarizer, or pulling a back catalogue as text. Spoken, Taddy, Podscript.
- Search across a transcript corpus for mentions: brand monitoring, guest research, PR. You want the episodes where a phrase appears, not one known episode. Podchaser, Pod Engine, Listen Notes.
- Generate a transcript from an audio file: unreleased recordings, or an episode published without a transcript. TranscriptFetch, podcasttranscript.ai, Deepgram, AssemblyAI, Whisper, Podsqueeze.
If you host a podcast, you are in the first group for anything you have already published. Fetching your own back catalogue as text is faster and cheaper than re-transcribing audio you already released. Group three is only for audio that isn't out yet, or episodes that shipped without a transcript.
Retrieval APIs: the transcript already exists
These hand back a transcript that was published with the episode. Seconds per episode, cents each, and speaker names when the published transcript carries them. The limit is coverage: an episode with no published transcript is a miss, not a wait. Spoken is listed first because this is Spoken's page; every entry says where that tool is the wrong pick.
Spoken — published transcripts with real speaker names
Returns the transcript that was already published with the episode as clean Markdown, with voices resolved to actual names, "Andrew Huberman" rather than "Speaker 1". One GET request, no audio handling, no diarization pass. Billing is per episode, so a three-hour interview costs exactly what a twenty-minute one does.
Best for: building a searchable archive or RAG corpus over published shows, where speaker attribution matters and the corpus gets reprocessed as models change.
"I use Spoken to feed an Obsidian vault where I collect all of my research input — PDFs, blog posts and now also podcasts."
Pick something else if: you need an episode that has no published transcript, which Spoken 404s on rather than generating, or you need YouTube, TikTok and Instagram in the same corpus.
Pricing: one-time credit packs, $15 / $50 / $160 for 100 / 500 / 2,000 transcripts ($0.15 down to $0.08 each). Credits never expire, errors are never charged, and re-fetching an episode you already pulled is free.
Taddy — podcast database with transcripts attached
A GraphQL API over 4 million shows and 200 million episodes, with a transcript query for any episode: if Taddy already holds the published transcript it returns that, and if not it generates one on demand, which its own docs put at ten seconds or more. Speaker names and timecodes come through when the published transcript has them. The rest of the API is directory data: search, charts, and webhooks for new episodes.
Best for: an app that needs podcast metadata, search and webhooks in one API, with transcripts as one field among many.
Pick something else if: transcripts are the whole job. Plans are priced on API requests, and the transcript allowance is the small number: 100 a month on Pro, 2,000 on Business, then packs of 3,000 for $75.
Pricing: Free plan with 500 requests a month, transcripts from the Pro plan up; Pro $75/mo with 100 transcripts; Business $150/mo with 2,000. Verified 22 September 2026.
Podscript — free, agent-native
A free, broader-source transcript tool built for agent use, with no credit model.
Best for: low-volume, exploratory, or hobby work where a free tool is sufficient.
Pick something else if: you're running production volume and need a support commitment behind it.
Apify actors — three small scrapers, pay per event
Apify's store carries three podcast-transcript actors. Two return the transcript a show published in its RSS feed, at $0.001 per transcript returned, with speaker labels when the publisher included them; the third runs Whisper inside the actor at $0.005 per episode plus a per-minute charge. All three publish their usage: 14, 2 and 1 users at the time of writing.
Best for: a pipeline already on Apify, where a transcript step slots in beside other actors.
Pick something else if: you need a track record behind the tool, or speaker names resolved rather than passed through.
Corpus search APIs: find the episodes that mention something
These index millions of transcripts and answer "which episodes said X". They will hand back a transcript too, but at database prices, and without speaker names.
Podchaser — transcript search across a very large catalogue
A podcast database with roughly 4 million episode transcripts behind REST and GraphQL, plus something the retrieval tools don't have: a single endpoint that searches every transcript in the catalogue and returns each episode where a term appears, filterable by audience size and brand-safety attributes. Also carries 27M host/guest/crew credits and chart data.
Best for: brand monitoring, PR, and media research. "Which episodes mentioned us last quarter" is a query the retrieval APIs cannot answer.
Pick something else if: you already know which episodes you want. You'd be paying database and monitoring pricing for a fetch.
Pricing: Starter $30/mo (1,000 requests, transcripts included), Professional $300/mo (10,000 requests), Enterprise custom with GraphQL and S3 bulk delivery.
Pod Engine — research and guest booking, with an API attached
Search 4M+ podcasts by charts, location, guests, reviews and genre, with transcript search included and both a REST API and an MCP server. Positioned around podcast research and guest booking rather than data extraction.
Best for: guest booking, podcast PR, and competitive research where the API is a convenience on top of a platform you're already using.
Pick something else if: you only want the API. The floor is a platform subscription, which is a lot of product to buy for a fetch endpoint.
Pricing: core platform from $75/mo; guest booking from $175/mo.
Listen Notes — the best-known podcast search API
The long-standing podcast search and directory API. Full episode transcripts are available through show_transcript=1, but only on paid tiers; the free tier returns metadata and no transcript text.
Best for: discovery and metadata at scale, where transcripts are a secondary need alongside search, charts and directory data.
Pick something else if: transcripts are the primary product you're buying. The entry point for transcript access is steep next to the retrieval tools.
Pricing: Free 300 requests/mo without transcripts; PRO $200/mo including 5,000 requests with transcripts, then $1.60 per extra 1,000; Enterprise custom.
Generate from audio: any episode, no names
These resolve a link to the audio and transcribe it. Coverage is anything with audio behind it, including the episodes retrieval tools miss; the trade is minutes instead of seconds, and diarization labels or nothing instead of names.
TranscriptFetch — broadest coverage, cheapest per transcript
Resolves an Apple, Spotify, YouTube or RSS link to the underlying audio and transcribes it. Because it generates rather than retrieves, coverage is effectively anything with audio behind the link, including the episodes Spoken 404s on. Ships a remote MCP server, SDKs in Python and Node, an n8n node, and a batch endpoint.
Best for: multi-platform corpora, or any job where coverage beats fidelity and cost per transcript is the deciding variable.
Pick something else if: you need speaker names; they ship no speaker identification, so the output has no attribution. Credits are also subscription-bound.
Pricing: monthly plans from $5, with 100 free credits a month and no card required. They are several times cheaper per transcript than Spoken at entry, and more at volume. If unit cost decides it for you, they win.
podcasttranscript.ai — machine transcription of a podcast link
Pass a Spotify link or an audio URL and it transcribes the episode with its own speech-to-text, returning text with timestamped segments and an AI summary. Speaker labels exist on its website but, per its own FAQ, are not returned by the API.
Best for: cheap transcription of episodes that have no published transcript, when timestamps matter more than who said what.
Pick something else if: you need speaker names in the API response, or the transcript that was actually published rather than a machine pass over the audio.
Pricing: 20 free transcriptions; $9/mo for 250, $19/mo for 1,000, $49/mo for 3,000; one-off packs of 100 for $7, 500 for $29 and 1,000 for $49 that never expire. Verified 22 September 2026.
TranscriptAPI — YouTube only
A clean, cheap transcript API with video search, channel browsing, playlist extraction, an MCP server and published agent skills. Its corpus is YouTube. There is no Apple or Spotify surface anywhere in its product.
Best for: video-first shows that publish full episodes to YouTube, which is a growing share of large podcasts. A real option, just not a podcast-native one.
Pick something else if: your shows are audio-first, or you need credits that survive a lapsed subscription; unused credits pause until you renew.
Pricing: $5/mo ($4.50 billed annually), 1,000 credits/month, top-ups $2.50 per 1,000, 100 free credits.
Supadata — web and YouTube to text
A general "web & YouTube to text" API aimed at makers, sold on credits with MCP support.
Best for: mixed pipelines pulling text from several kinds of source through one vendor.
Pick something else if: podcasts are the whole job; a podcast-native API will handle episode and show semantics that a general web-to-text tool won't.
Feature comparison
The row that decides most evaluations is the billing unit. Per-episode and per-minute pricing diverge sharply on long-form podcasts, which routinely run two to three hours.
| Spoken | TranscriptFetch | Podchaser | Listen Notes | Taddy | podcasttranscript.ai | Deepgram / AssemblyAI | |
|---|---|---|---|---|---|---|---|
| Billing unit | Per episode | Per transcript | Per API request | Per API request | Per request, transcript allowance | Per transcription | Per minute of audio |
| 3-hour episode costs | Same as a 20-minute one | Same as a 20-minute one | 1 request | 1 request | 1 transcript | Same as a 20-minute one | 9x a 20-minute one |
| Commitment | One-time packs | Monthly subscription | Monthly subscription | Monthly subscription | Monthly subscription | Subscription or one-off packs | Usage / committed spend |
| Credits expire | Never | With the subscription | Monthly quota | Monthly quota | Monthly quota | Quota monthly; packs never | n/a |
| Transcript source | Published with the episode | Generated from audio | Catalogue | Catalogue | Published, else generated | Generated from audio | Generated from your audio |
| Real speaker names | Yes, resolved per turn | No | No | No | If the published transcript has them | Not via the API | Diarization labels only |
| Output format | Markdown | JSON / text | JSON | JSON / text | JSON | JSON | JSON |
| Search a corpus for a phrase | No | No | Yes | Yes | No | No | No |
| Re-fetch cost | Free, forever | Charged | Charged | Charged | Charged | Charged | Charged |
| Entry price | $15 one-time | $5/mo | $30/mo | $200/mo for transcripts | $75/mo | $9/mo, or a $7 pack | Per-minute rates |
Why per-episode vs per-minute is the decisive difference
Speech-to-text vendors bill by audio duration. Podcasts are long: a two-hour episode is unremarkable, and interview shows regularly run three. On a per-minute vendor, archiving a 300-episode back catalogue at an average 90 minutes means paying for 27,000 minutes of audio, and the bill scales with how talkative the show is.
On Spoken that same archive is 300 credits, $24 at the 2,000-pack rate, because the unit is the episode. The per-minute route also costs you the audio download, a diarization pass, and a second LLM call to turn "Speaker 0" into a name.
Full breakdowns: vs Deepgram, vs AssemblyAI, vs Whisper + Pyannote.
Tools that appear on these lists and shouldn't
Three names come up constantly for this query and solve a different problem.
- Podsqueeze: a creator tool for your own show, turning episodes into show notes, clips and social posts. It works from audio you upload, so even for your own back catalogue it re-transcribes episodes you've already released rather than fetching the transcript that already exists. And there's no path to any show you don't publish.
- Deepgram and AssemblyAI: excellent speech-to-text APIs for audio you hold. For a published podcast they mean locating and downloading 50–100 MB of audio first, then paying per minute to regenerate a transcript that already exists.
- Whisper: the same shape as above, self-hosted. Free per se, but you're paying for the GPU time, the diarization pipeline and the speaker-naming pass.
How to choose in one pass
| If your job is… | Use |
|---|---|
| An archive or RAG corpus of published shows, with speaker attribution | Spoken |
| Maximum coverage across podcasts, YouTube and social, cheapest per unit | TranscriptFetch |
| Finding every episode that mentions a brand or phrase | Podchaser |
| Booking guests or podcast PR research | Pod Engine |
| Podcast discovery and metadata, transcripts secondary | Listen Notes |
| Video-first shows published to YouTube | TranscriptAPI |
| Podcast metadata, search and webhooks in one API, transcripts included | Taddy |
| Cheapest machine transcription of a podcast link, timestamps without names | podcasttranscript.ai or TranscriptFetch |
| Transcribing audio you recorded yourself | Deepgram, AssemblyAI, or Whisper |
| Turning your own show into show notes and clips | Podsqueeze |
| Getting your own published back catalogue back as text | Spoken |
Try Spoken without signing up
curl -H "x-api-key: pt_demo" \
https://spoken.md/transcripts/1000651996090For the full loop over a show's back catalogue in Python, Node and Bash, see how to get podcast transcripts programmatically.
FAQ
What is the best podcast transcript API for AI agents?
The deciding factors for agent use are a callable tool interface and output an LLM can consume without cleanup. Spoken returns Markdown with speaker turns already bolded, and ships an MCP server two ways, a hosted endpoint at https://spoken.md/mcp and spoken-mcp on npm and in the official MCP registry, so an agent can call it as a tool without you writing a client. TranscriptFetch, TranscriptAPI and Pod Engine also ship MCP servers; TranscriptFetch's and Pod Engine's are remote endpoints too.
Which podcast transcript API is best for building a knowledge base?
Three things matter for a knowledge base that outlives its first build: attribution, format, and reprocessing cost. Speaker names make a retrieved chunk answerable ("who said this?") instead of anonymous. Markdown chunks on speaker boundaries with a standard splitter, no parsing layer. And because you will reprocess the corpus when your embedding model or chunking strategy changes, a vendor that charges again for the same episode turns a one-time cost into a recurring one. Spoken's re-fetches are free and its credits don't expire. See RAG over podcasts.
What is the best podcast data API for LLMs?
Split the question: if you need transcript text to feed a model, use a retrieval API (Spoken, TranscriptFetch). If you need structured metadata such as charts, credits, audience size and reviews, use a database API (Podchaser, Listen Notes). Many production systems use one of each, because neither category covers the other well.
How do I get podcast transcripts as Markdown?
Spoken returns text/markdown directly: bold speaker turns, parenthetical timestamps, UTF-8, paragraph breaks following the natural pauses in conversation rather than a fixed line length. Most other APIs on this list return JSON, which you'd convert yourself.
Is there a free podcast transcript API?
Several podcast transcript APIs have free entry points, and they differ in kind. Podscript is free outright. TranscriptFetch gives 100 credits a month with no card. TranscriptAPI includes 100 free credits. podcasttranscript.ai gives 20 free transcriptions. Taddy's free plan covers 500 requests a month, and transcripts start on its $75 Pro plan. Listen Notes' free tier explicitly excludes transcript text. Spoken has no recurring free tier but pt_demo searches and lists any show and fetches the demo episode without signup, and both search endpoints are free on any key.
Can I get transcripts for my own podcast?
Yes, a retrieval API returns your own podcast's transcripts, and it's usually the cheapest route. If your episodes are already published with transcripts, a retrieval API hands them straight back as Markdown, with no upload and no per-minute charge. That applies whether or not you host the show, so podcasters use it to pull their own back catalogue for show notes, search and repurposing. You only need a speech-to-text tool for audio that isn't published yet, or episodes that went out without a transcript.
Can I get a transcript for any podcast episode?
No single vendor returns a transcript for every podcast episode. Tools that retrieve published transcripts are limited to episodes that have one. Spoken returns a 404 rather than a guess, and its search endpoints only return episodes that are actually fetchable. Tools that generate from audio, like TranscriptFetch, cover far more but produce a machine transcription without speaker names. That trade between coverage and fidelity is the real choice.
What does a podcast transcript cost per hour of audio?
The cost of a podcast transcript per hour depends entirely on the billing unit. Per-minute vendors charge by duration, so an hour costs 60 minutes of billing regardless of what's in it. Spoken charges per episode, from $0.08 to $0.15, so an hour and three hours cost the same.
TL;DR: If you're retrieving transcripts for published shows and need real speaker names in Markdown, Spoken is the one built for that and costs the same whether the episode runs 20 minutes or three hours. If you need maximum coverage at the lowest unit price, TranscriptFetch beats it. If you need to search a corpus rather than fetch known episodes, you want Podchaser or Listen Notes, not a retrieval API at all.
Thousands of transcripts fetched by developers building summarizers, RAG pipelines, and podcast tools