This is a clone of the Modal Podcast Transcriber
This is a complete application that uses OpenAI Whisper to transcribe podcasts. Modal spins up 100-300 containers for a single transcription run, so hours of audio can be transcribed on-demand in a few minutes.
- Modal.com LIVE App: You can find the app here: https://modal-labs--whisper-pod-transcriber-fastapi-app.modal.run/
- Oncetold App: https://gagglepod--whisper-pod-transcriber-fastapi-app.modal.run
The entire application is hosted serverlessly on Modal and consists of 3 components:
- React + Vite SPA (
pod_transcriber/frontend/) - FastAPI server (
pod_transcriber/api.py) - Modal async job queue (
pod_transcriber/main.py)
accountapproved Modal.com accountnpmmodalinstalled in your current Python virtual environment
To run this on your own Modal account, you'll need to create a Podchaser account and create an API key.
Once you have the Podchaser account and API keys, then, create a Modal Secret with the following keys:
PODCHASER_CLIENT_SECRETPODCHASER_CLIENT_ID
It doesn't matter what you call the Modal Secret block -- what matters is that both KEYS (with VALUES) are listed in the block (Note: This will not work locally).
You can find both on their API page.
cd into the pod_transcriber/frontend directory, and run:
npm installnpx vite build --watch
The last command will start a watcher process that will rebuild your static frontend files whenever you make changes to the frontend code.
Once you have vite build running, in a separate shell run this at the app root to start an ephemeral app on Modal:
modal serve oncetold-podcast-transcriber.mainPressing Ctrl+C will stop your app.
Once your happy with your changes, cd to the root of the project and run modal deploy oncetold-podcast-transcriber.main to deploy your app to Modal.
Modal.com deployment allows you to transcribe HTML only at about $.15/60-minutes of audio. However, you can cut and paste what looks like an SRT version of the transcript to your own *.SRT file.