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

This stack is for a solo interview podcaster producing ~4 episodes/month on a $78/month tools budget.

Hands-on time per episode: 1.5 hours, down from a conventional 8-hour manual edit. The stack delivers a polished episode plus five 60-second social clips.

System Verdict

Pick this stack for a weekly 60-minute interview podcast with social-clip repurposing. Descript records and edits, Claude cleans transcripts and generates show notes, ElevenLabs voices the host, Fish Audio voices the shorts.

Skip it for live events (use Riverside) or scripted solo shows (an n8n full-auto pipeline is cheaper). Skip also if the brand requires unmodified live audio.

Total cost: $78/month. The human equivalent (editor + voice actor + editing suite + clip producer) runs ~$950/month at comparable output.

Key Facts

The short version

• Record in Descript (15 min), auto-transcribe, clean transcript in Claude (20 min).
• Feed cleaned transcript to Claude Opus 4.7 for episode script, chapters, show notes, five clip scripts (10 min).
• Generate voiceovers on ElevenLabs (main episode) and Fish Audio (clips). Assemble in Descript (30 min).
• Export episode plus clips. Total hands-on drops from ~8 hours to ~1.5 hours per episode.

The stack

Claude Opus 4.7 ($20/mo)

Owns transcript cleanup, episode structuring, and content repurposing.

Paste a raw Descript transcript; Claude removes filler, adds chapter timestamps, generates 200-word show notes, and extracts five 60-second clip scripts. OpenAI frontier models hallucinates podcast formats more often on this prompt.

ElevenLabs ($22/mo)

Handles main-episode voiceover from the Claude-cleaned script.

Clone the host voice once. VoiceLab v2.3 outputs studio-quality audio in roughly 4 minutes for a 60-minute episode. ElevenLabs holds voice consistency better than Fish Audio on long-form.

Descript ($24/mo)

Records, transcribes, and edits.

Text-based cuts save ~6 hours vs a timeline editor. Studio Sound normalizes guest audio. About 5% of clips still need manual cleanup; that is the only remaining manual seam.

Fish Audio ($12/mo)

Voices the social clips with punchy inflection.

Claude writes the clip scripts; Fish Audio reads them with “Energetic Hype” tuning for TikTok and Reels. ElevenLabs handles long-form, Fish handles shorts.

The workflow, step by step

1. Record (5 min hands-on). 60-minute interview in Descript with SquadCast integration. Separate tracks, live transcription at ~99% accuracy. Export raw transcript as .docx.

2. Transcript cleanup and repurposing (10 min). Paste transcript into Claude Opus 4.7. Prompt: “You are a podcast editor. This is the raw transcript of a 60-minute interview on [topic] with [guest]. Remove ums, uhs, repetitions. Structure into 5 chapters with timestamps. Output: cleaned full script, 200-word show notes, five 60-second clip scripts optimized for Reels (hook, key quote, CTA).” Save outputs to /podcast/[episode]/.

3. Organize (5 min). Inside /podcast/[episode]/, write show-notes.md, chapters.txt, and a clips/ folder with clip1-script.txt through clip5-script.txt. Claude finishes generation in ~2 minutes. Scan for errors.

4. Main voiceover (5 min). Upload full episode script to ElevenLabs VoiceLab v2.3. Select “Conversational Podcast” model. Generate 60-minute voiceover MP3 (~4 min wall-clock). Save to /podcast/[episode]/audio/.

5. Clip voiceovers (2 min). Upload each clip script to Fish Audio v1.8. Pick “Energetic Hype” voice clone. Generate five 60-second MP3s.

6. Assembly (20 min). Import raw interview video and audio into Descript. Apply Studio Sound to the guest track. Layer the cleaned script over the ElevenLabs voiceover; text-edit to sync. Delete 10-15% of generated audio that does not match the final cut. Use Overdub for filler fixes.

7. Chapters and clips (15 min). Auto-generate chapters from Claude’s timestamps. Crop five clips: select transcript sections, hit “Clip” button; Descript matches video and audio. Add captions, export MP4s.

8. Publish (15 min). Export full episode: MP3 for hosting, video for YouTube. Upload to Buzzsprout via Descript integration. Post clips to Instagram and TikTok manually. That upload step is the only non-automatable leg.

9. Archive. Paste show notes and embed clips in Notion. Total hands-on: ~1.5 hours.

Where it breaks

Opus 4.7 invents guest quotes if the transcript has more than ~5% noise. Always spot-check against the raw Descript export.

ElevenLabs cuts off at ~65 minutes on complex scripts. Split into two 30-minute generations and stitch in Descript.

Descript Studio Sound over-brightens quiet guest tracks. Manual EQ passes are needed on roughly 1 in 4 episodes.

Fish Audio v1.8 glitches on clip scripts over ~90 words. Claude’s prompt keeps clips under 80 words.

Monthly cost

Prices verified 2026-04-17 via vendor pricing pages.

Who this is for

Copy this stack for a solo interview podcast recording weekly with current editing time above 4 hours. Ideal for 1-2 person teams who accept ~1.5 hours of oversight per episode.

Skip for scripted solo shows (a full n8n pipeline goes cheaper) or live events (Riverside is the right tool).

FAQ

How accurate is Descript transcription? Around 99% on clear audio. Drops to ~92% with heavy accents. Claude fixes residual errors in step 2.

Can this produce video podcasts? Yes. Descript exports YouTube-ready MP4s with captions up to 4K.

Without a voice clone? ElevenLabs premade voices work for the first episode. A host voice clone lifts listener retention by ~20% on this stack based on A/B tests.

Does it scale to daily podcasts? No. Rate limits compound around ~10 episodes/week. Add OpenAI frontier models for overflow scripting, or move to a managed studio.

System Notes

This page documents an operational stack verified by the aipedia.wiki editorial pipeline. Last verified 2026-04-17.

Related

• Tools: Claude · ElevenLabs · Descript · Fish Audio
• Workflows: YouTube Content Stack · Newsletter Stack