How do I search podcasts by who is speaking?

To search podcasts by who is speaking, you need transcripts where every segment is labeled with a named speaker, not just “Speaker 1”. Audioscrape indexes more than 18 million public episodes that way, so you can type speaker: "Jerome Powell" rate cuts into the search bar and get only the moments where that person said those words, each with a timestamp and a link to the audio. The same filter is available to AI assistants over MCP and to scripts over the REST API.

Why this is hard without a speaker-attributed index

Podcast apps search titles and show notes, not the spoken words. Raw transcription services give you text with anonymous speaker labels that reset on every file, so “Speaker 2” in one episode has nothing to do with “Speaker 2” in the next. To answer “what did this person say”, someone has to attach a real name to each voice and keep that name consistent across episodes and shows. That is the part Audioscrape does for the public corpus.

How to do it with Audioscrape

On the website (no account needed)

1. Go to audioscrape.com/search.
2. Put the speaker filter directly in the query, for example speaker: "Elon Musk" starship. Quoted phrases match exactly; you can combine the speaker filter with podcast: "Show Name", a date range like from: 2024 to: 2024-06, and boolean operators (AND, OR, NOT).
3. Or run a plain search first and use the Speakers facet in the sidebar. It lists who said matching things and how many matches each has; click a name to narrow the results.
4. Every result shows the speaker, the transcript snippet with your terms highlighted, the podcast and episode, and a clickable timestamp that jumps to that second of audio.

You can also start from a person: search their name, open their page at /person/{slug}, and browse every attributed appearance.

From Claude, ChatGPT, Cursor or another MCP client

Connect the Audioscrape MCP server at https://mcp.audioscrape.com (setup guides for Claude Desktop, ChatGPT and Cursor). Then ask in plain language, for example “Find what Sam Altman said about AGI in the last six months, with timestamps”. Under the hood the assistant calls:

• search_speakers to resolve a name to a person_slug,
• get_speaker for that person’s bio and appearances,
• search_audio for the actual quotes, with search_type set to text for exact wording or semantic for the idea.

From code

POST https://www.audioscrape.com/api/search with a Bearer API key. Pass "filters": {"speakers": ["Dr. Jane Smith"]} alongside your query. Set "options": {"include_aggregations": true} to get a speakers bucket list back, which is handy for discovering who talks about a topic before you filter. Details and response shape: Search API.

What you get

• Speaker on every segment, with a slug that links to the person page.
• Timestamps (start_timestamp, end_timestamp) and a segment URL that opens the episode at that moment.
• Episode and podcast metadata: title, publish date, publisher, category.
• Entities mentioned in the segment (people, companies, topics).
• Website search needs no account. MCP and API access need a free account and share one quota of 100 searches per month; paid plans raise that, and semantic search is a Pro and Enterprise feature.

Limits and honest caveats

• Speaker names in the public corpus are attributed automatically by an evidence-gated classifier. It is good but not perfect; guests on small shows are the most likely to be missing or mislabeled. If the speaker filter does not show the name you expect, search the name as plain text instead. Mis-attributions can be reported from any person page, and corrections are queued for review.
• Search returns segments, not whole episodes. Use get_transcript (MCP) or the episode page for the full transcript.
• Not every podcast is indexed. If a show is missing, submit its RSS feed at /submit.
• Transcription is batch, not live, so an episode published an hour ago may not be searchable yet.

Doing it without Audioscrape

If the episode is short and you know which one you want, you can download it from the show’s own feed, run it through a transcription tool that supports diarization, and read the anonymous speaker labels against the intro to work out who is who. That works for one episode. It does not scale to “everything this person has said across 40 shows”, which is what a speaker-attributed index is for.

Plans and limits: /pricing. Facts for AI agents: /docs/for-agents.