Eight APIs that return podcast transcripts, ranked for the job most people mean — pulling transcripts for episodes that are already published, programmatically.
Last updated August 2026 — prices verified against each vendor's own pricing page
"Podcast transcript API" means three different jobs, and most comparison lists mix them. Before the ranking, work out which one you're doing — it eliminates most of this list immediately.
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 — you do not need to hand your episodes to a transcription service to get them back as Markdown. Group three is only for audio that isn't out yet, or episodes that shipped without a transcript.
This page ranks the first group, then covers the other two honestly so you can rule them in or out. If you're in group three, none of the top five will help you and that's not a criticism of them.
Ranked for one job: fetching transcripts for episodes that are already published, through an API, at volume. A tool that's badly suited to that may be excellent at something else, and where that's true it says so.
Returns the transcript that was already published with the episode as clean Markdown, with voices resolved to actual names — "Andrew Huberman", not "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.
Pick something else if: you need an episode that has no published transcript — Spoken returns a 404 rather than generating one — or you need YouTube, TikTok and Instagram in the same corpus. Coverage is the honest trade for fidelity.
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.
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 feature — the output is machine transcription without attribution. Also note credits are subscription-bound.
Pricing: monthly plans from $5, with 100 free credits a month and no card required. They are meaningfully cheaper per transcript than Spoken — several times over at entry, more at volume. If unit cost is what decides this for you, they win it, and no framing on this page changes that.
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 simply 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.
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.
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.
A clean, cheap transcript API with video search, channel browsing, playlist extraction, an MCP server and published agent skills. It's on this list because it's constantly recommended for podcasts, so it's worth being precise: 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 an increasing share of large podcasts, so this is 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.
A general "web & YouTube to text" API aimed at makers, sold on credits with MCP support. Podcasts aren't the focus, but it belongs here as a competent option if podcast transcripts are one input among many in a broader scraping pipeline.
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.
A free, broader-source transcript tool built for agent use, with no credit model. Worth trying precisely because it costs nothing to evaluate.
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.
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 | Deepgram / AssemblyAI | |
|---|---|---|---|---|---|
| Billing unit | Per episode | Per transcript | Per API request | Per API request | Per minute of audio |
| 3-hour episode costs | Same as a 20-minute one | Same as a 20-minute one | 1 request | 1 request | 9x a 20-minute one |
| Commitment | One-time packs | Monthly subscription | Monthly subscription | Monthly subscription | Usage / committed spend |
| Credits expire | Never | With the subscription | Monthly quota | Monthly quota | n/a |
| Transcript source | Published with the episode | Generated from audio | Catalogue | Catalogue | Generated from your audio |
| Real speaker names | Yes, resolved per turn | No | No | No | Diarization only — "Speaker 0" |
| Output format | Markdown | JSON / text | JSON | JSON | JSON |
| Search a corpus for a phrase | No | No | Yes | Yes | No |
| Re-fetch cost | Free, forever | Charged | Charged | Charged | Charged |
| Entry price | $15 one-time | $5/mo | $30/mo | $200/mo for transcripts | Per-minute rates |
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. That's before the per-minute route 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.
Three names come up constantly for this query and solve a genuinely different problem. They're good products — they just won't do this job.
| 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 |
| 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 |
curl -H "x-api-key: pt_demo" \
https://spoken.md/transcripts/1000651996090pt_demo is a real key. It works on every endpoint without signup, so you can check the output shape before paying for anything.
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 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, which is easier to wire into a hosted agent than a stdio server.
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.
Split the question: if you need transcript text to feed a model, use a retrieval API (Spoken, TranscriptFetch). If you need structured metadata — charts, credits, audience size, reviews — use a database API (Podchaser, Listen Notes). Many production systems use one of each, because neither category covers the other well.
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.
Several 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. Listen Notes' free tier explicitly excludes transcript text. Spoken has no recurring free tier but pt_demo works on every endpoint without signup, and both search endpoints are free on any key.
Yes, and it's usually the cheapest route. If your episodes are already published with transcripts, a retrieval API hands them straight back as Markdown — no upload, no per-minute charge, no transcription job to wait on. 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.
Not from any single vendor. 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 — coverage against fidelity — is the real choice on this page.
It 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 top pick 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
No signup required — use API key pt_demo on any endpoint.