I run a podcast for a Vancouver-based marketing consultancy. Every week we record an hour-long conversation with a founder or agency owner. For the first six months, I paid $29/month for a transcription service that would give me back a messy text file with speaker labels that were wrong half the time. Then I'd spend another 45 minutes cleaning it up, formatting show notes, and pulling out quote cards for social.
Two months ago I built a Claude Code podcast transcription automation that does the whole thing in under 10 minutes. No monthly fee. No manual cleanup. Just drop the MP3 file into a folder and get back a formatted transcript with timestamps, chapter markers, and five social media quote cards ready to post.
Here's exactly how it works, what it costs, and how you can build the same thing if you're producing regular podcast content.
The Problem with Existing Transcription Tools
If you've tried any of the popular podcast transcription services — Otter, Descript, Rev — you know they're fine for one-off episodes. But they have three problems when you're running a regular show:
- Monthly subscription lock-in — you're paying $20–$50/month whether you publish one episode or ten
- Manual post-processing — the transcript comes back with speaker labels like "Speaker 1" and "Speaker 2," and you still have to format everything for your website or YouTube description
- No workflow integration — you can't automatically pull out the best quotes, generate chapter markers, or format everything exactly how your editor needs it
The real cost isn't the subscription. It's the 30–60 minutes you spend after every episode fixing formatting, adding timestamps, and turning the raw transcript into something usable. That time adds up fast when you're doing weekly or bi-weekly episodes.
How I Automated Podcast Transcription with Claude Code
The workflow I built runs entirely locally. No third-party API calls, no usage limits, no recurring fees. It uses Claude Code to handle the audio-to-text conversion and the formatting logic in a single script.
Here's what happens when I drop a new episode file into the input folder:
- Claude Code transcribes the audio using Whisper (OpenAI's open-source transcription model, which runs locally)
- It identifies speakers based on voice patterns and labels them correctly (I provide a config file with host and guest names)
- It generates timestamps for every speaker turn
- It pulls out 5–7 key moments and formats them as chapter markers with descriptions
- It extracts five quotable moments and formats them as social media cards (text + timestamp)
- It outputs three files: full transcript (Markdown), show notes (HTML), and social clips (JSON)
Total runtime for a 60-minute episode: about 8 minutes on my MacBook Pro. The transcript quality is better than the subscription tools I was using before, and I don't have to do any manual cleanup unless there's a technical term that needs correction.
The Technical Setup
If you want to replicate this, here's what you need:
- Claude Code (free for personal use, or $20/month for the Pro version if you're doing high volume)
- Whisper installed locally (open-source, no API cost)
- A config file with your show name, host name, and typical guest name format
- A folder structure: /input/ for raw audio, /output/ for transcripts, /archive/ for processed files
The first time you set it up takes about two hours. After that, it's fire-and-forget. I haven't touched the script in six weeks — it just runs automatically whenever I drop a new file in the folder.
What the Output Looks Like
The formatted transcript includes proper speaker labels, paragraph breaks at natural pauses, and timestamps every 30 seconds. It looks like this:
[00:00:12] Alejandro Arce: Welcome back to the show. Today we're talking about how B2B companies are using AI to accelerate content production without losing brand voice. [00:00:19] Guest Name: Thanks for having me. This is a topic we've been living and breathing for the past year, so I'm excited to dig in.
The chapter markers come out in a format that works for YouTube descriptions, Apple Podcasts chapters, or any other platform that supports them:
00:00 – Intro and context 02:15 – Why most AI content feels generic 08:40 – The three-layer content system 15:22 – Case study: 10x content output in 90 days 22:05 – Common mistakes to avoid
And the social clips are delivered as JSON so I can pipe them directly into a social media automation workflow:
{
"quote": "We went from publishing 2 articles a month to 20, and quality went up.",
"timestamp": "15:47",
"speaker": "Guest Name"
}
My editor takes those five clips, drops them into Canva templates, and we have a week of LinkedIn and Twitter content ready to go. Total time from raw audio to published social posts: about 20 minutes, and most of that is my editor's visual QA.
Cost Comparison: Subscription vs. Automation
Let's break down what this actually costs over a year of weekly podcast production:
Otter.ai Business plan: $30/month × 12 months = $360/year
Manual formatting time: 45 min/episode × 50 episodes × $50/hour = $1,875/year
Total annual cost: $2,235
Claude Code automation: one-time build cost of ~$500 (if you hire someone) or 2 hours of your time
Ongoing cost: $0 (runs locally, no API fees)
Total annual cost: $500 first year, $0 after that
Even if you only produce 12 episodes a year, you break even in the first year. If you're doing weekly content like we are, the ROI is immediate.
When You Shouldn't Automate Podcast Transcription
This approach isn't for everyone. If any of these apply to you, stick with a manual service:
- You only publish 1–2 episodes per year (the setup time isn't worth it)
- Your episodes have heavy accents, overlapping speakers, or poor audio quality (automated transcription will struggle)
- You need legally defensible transcripts for regulatory compliance (human review is still the standard)
- You don't have 2 hours to set up the initial workflow
But if you're publishing regularly, have decent audio quality, and you're tired of paying for the same task every month, this is one of the highest-ROI automations you can build.
How to Get Started
If you want to build this for your own podcast, here's the order I'd recommend:
- Start with basic Claude Code setup and get comfortable running scripts
- Install Whisper locally and test it on one episode manually to confirm transcription quality
- Build the speaker identification logic (this is the trickiest part — start simple and refine over time)
- Add the formatting layer (timestamps, chapters, social clips)
- Run it on your last 3 episodes as a test before committing fully
The whole process took me about 6 hours spread over two days, and I'm not a developer. If you know Python or have done any workflow automation before, you can probably do it faster.
If you'd rather have someone build it for you, that's exactly the kind of project I take on for Vancouver clients. Reach out and we can scope it in a 20-minute call. I can usually deliver a working system within a week.
What This Unlocks Beyond Transcription
The biggest benefit isn't the time savings on transcription — it's what becomes possible once you have structured, formatted transcripts at scale.
We're now using the transcripts to:
- Auto-generate SEO-optimized blog posts from each episode (another Claude Code workflow)
- Build a searchable podcast archive on the website with full-text search
- Create email newsletter content by pulling the top three insights from each episode
- Feed episode data into our CRM so the sales team knows which topics a prospect has engaged with
None of that was realistic when transcription was a 60-minute manual task. Now it's just part of the automated pipeline. One episode recording becomes 8–10 content assets without any additional human time.
If you're already producing regular podcast content, this is low-hanging fruit. The tools are there, the setup is straightforward, and the payoff is immediate. You just have to decide whether you want to keep paying the subscription tax or spend a few hours building something you own forever.
Questions about how this would work for your specific show format? The FAQ page covers most of the edge cases I've run into. And if you want to see the actual workflow in action, book a call and I'll walk you through it live.