Podcast Automation Stack: Claude, ElevenLabs, Descript, Fish Audio

This stack is for a solo or small-team interview podcast that wants one repeatable workflow for transcript cleanup, episode structure, show notes, optional voiceover, and short-form clips.

AiPedia verdict, verified June 12, 2026: use Descript as the recording, transcript, edit, and clip workspace; use Claude for transcript cleanup, show notes, chapters, and repurposing; use ElevenLabs for high-quality voice work when the host has consented to cloning or synthetic narration; and use Fish Audio for lower-cost short-form voice experiments. Keep Riverside on the shortlist when live-event recording quality matters more than text-based editing, and use Castmagic or MeetGeek when all you need is transcript-to-notes output.

The stack can reduce editing time, but it should not hide synthetic voice use. If a generated or cloned voice appears in the episode or clip, disclose it where listeners make trust decisions.

The Short Version

• Record and transcribe in Descript.
• Clean the transcript and create chapters, show notes, titles, descriptions, and clip scripts in Claude.
• Edit the main episode in Descript using text-based cuts, Studio Sound, Remove Filler Words, and Underlord where helpful.
• Use ElevenLabs only for consented host clones, intros, ad reads, narration, or replacement lines that listeners would not mistake for untouched live audio.
• Use Fish Audio for lower-cost short-form variants, test reads, and clip narration when the voice rights are clear.
• Publish with disclosure, source links, and a manual review of every claim.

The Stack

Claude: transcript cleanup and editorial structure

Claude owns the text work: transcript cleanup, chapter summaries, episode description, title options, guest bio, newsletter copy, and short-form clip scripts. Claude Pro is currently listed at $20/month when billed monthly, with Claude Code and Cowork included, but podcast teams should budget around general Claude usage rather than assuming every long transcript will fit comfortably into a light plan.

Use Claude for organization and drafting. Do not let it invent guest quotes, sponsor claims, statistics, or legal/medical advice.

Descript: recording, transcript editing, and clips

Descript is the production workspace. Its current product navigation includes podcasting, Rooms, captions, transcription, AI speech, Create Clips, Studio Sound, Remove Filler Words, and Underlord. The pricing page lists Free, Creator, Business, and Enterprise paths, with Creator showing 10 media hours/month and 400 AI credits/month, and Business showing larger media-hour and AI-credit allowances.

The important cost detail is metering. Descript’s help docs say media minutes are consumed by uploads and recordings, while AI credits are consumed by AI-powered features such as Underlord, Studio Sound, Remove Filler Words, Green Screen, Eye Contact, AI speech, avatars, and generated video. Some features scale with media length, and top-ups are available from the usage tab.

ElevenLabs: premium synthetic voice

ElevenLabs is the higher-quality voice lane. The current pricing page lists Creator at $22/month, with a first-month discount displayed, Professional Voice Cloning, and 121k credits/month. Pro, Scale, Business, and Enterprise expand credits, quality, seats, clones, concurrency, and business controls.

Use ElevenLabs for voiceover, intro/outro variants, ad reads, localization, or carefully disclosed voice-clone work. Keep the original speaker’s consent and review path in writing.

Fish Audio: budget voice variants and shorts

Fish Audio is the budget voice lane. The current plan page lists a Free tier, Plus, Pro, Max, and Enterprise. Plus is shown at $11/month when billed annually, with 250,000 credits/month, up to 200 minutes generation, larger character limits, private voice slots, priority generation, enhanced voice cloning, and commercial use allowed. Pro and Max increase credits, minutes, seats, and production capacity.

Use Fish Audio for short-form tests and clips. Do not rely on it for a full host-voice replacement without careful quality review and consent.

Workflow, Step By Step

1. Record the interview. Use Descript Rooms or your preferred recorder. Capture separate tracks when possible and keep the raw recording archived.

2. Create the transcript. Let Descript transcribe the episode. Before using AI cleanup, scan for speaker labels, names, brand terms, sponsor language, medical/legal claims, and anything that might become a quote.

3. Clean and structure in Claude. Paste the transcript or selected sections into Claude with instructions: remove filler without changing meaning, flag unclear sections, create chapter titles, draft show notes, extract quotes only from the transcript, and propose five clip candidates.

4. Edit in Descript. Make text-based cuts, apply Studio Sound carefully, remove filler words only when it does not change speaker intent, and create clips from moments that actually occurred in the interview.

5. Generate optional voice assets. Use ElevenLabs for premium narration or host clone work only with consent. Use Fish Audio for short-form alternatives or quick test reads. Label any synthetic or cloned voice in the production notes and public description where appropriate.

6. Review claims and quotes. Compare the final episode description, show notes, sponsor claims, and clip captions against the transcript. Do not publish Claude-generated claims without source checks.

7. Export and publish. Export the full episode audio/video, captions, and short clips. Keep the transcript, prompts, generated copy, voice files, and approvals in one folder.

8. Archive the episode package. Store /raw, /transcript, /claude-notes, /descript-project, /voice, /clips, /exports, and /approval folders. This keeps future corrections and repurposing sane.

Where It Breaks

Claude can over-clean transcripts and make a guest sound more certain than they were. Ask it to preserve meaning and flag unclear sections instead of silently rewriting them.

Descript AI credits and media minutes can run out faster than expected when the team uses Underlord, Studio Sound, speech regeneration, clips, avatars, or generated video on every episode.

ElevenLabs sounds good enough that disclosure matters. A cloned host voice used for ad reads or replacement lines should be approved by the speaker and disclosed when it affects listener trust.

Fish Audio is attractive on price, but short clips still need pronunciation, pacing, and rights review. Budget voice output can create brand risk if it sounds like a fake testimonial or fake guest quote.

Social clips are not automatically safer because they are short. A misleading 30-second quote can cause more damage than a full episode with context.

Monthly Cost

Treat the total as a monthly range. The cheapest honest version uses annual-rate self-serve plans and limited AI credit use. The professional version adds media-minute top-ups, voice-credit top-ups, team seats, or a higher ElevenLabs/Fish Audio tier.

Who This Is For

Copy this stack if:

• you publish interview episodes regularly,
• you already review audio and copy before publishing,
• show notes and clips are a meaningful growth channel,
• synthetic voice use is optional or explicitly disclosed,
• the time saved is worth more than the credit spend.

Skip it if:

• the brand requires untouched live audio,
• you cannot review transcript accuracy,
• the show covers high-stakes legal, medical, or financial advice without expert review,
• the host or guest has not consented to voice cloning,
• the team only needs show notes and no editing workflow.

System Notes

This page documents an operational podcast-production stack verified by the aipedia.wiki editorial pipeline. Last verified 2026-06-12.