title	Podcast Transcribe
emoji	🎙️
colorFrom	blue
colorTo	green
sdk	docker
app_port	7860

Podcast Transcribe

A web app for podcasters: load an audio file, transcribe it locally with OpenAI Whisper, then edit the transcript segment by segment while it stays synchronized with the audio.

Features

• Audio player with a real waveform drawn from the loaded file (any audio/* file).
• Transcription with local openai-whisper through a Python script; no API key needed.
• Time-stamped segments (start, end, text) shown next to the player, with the text editable in place.
• Export of the edited transcript as .txt or .json.

How it works

The Next.js page uploads the audio to POST /api/transcribe-local. The route saves it to a temporary file, runs scripts/transcribe_local.py, and returns { "segments": [{ "start", "end", "text" }] }.

The repo also contains POST /api/transcribe, which sends the file to the OpenAI Whisper API (needs OPENAI_API_KEY). The current UI does not call it.

Run locally

Requirements: Node.js 20+, Python 3.10+, and ffmpeg on your PATH.

# 1. Python environment with Whisper (the app uses ./.venv automatically if it exists)
uv venv
uv pip install -r scripts/requirements.txt

# 2. Web app
npm install
npm run dev

Open http://localhost:3000. The first transcription downloads the Whisper model (base by default).

Environment variables (optional):

Variable	Purpose	Default
WHISPER_MODEL	Whisper model: tiny, base, small, medium, large, turbo	base
WHISPER_PYTHON	Python command used to run the script	./.venv Python, else python3
OPENAI_API_KEY	Only for the unused /api/transcribe route	none

More detail on the local setup: scripts/README.md.

Docker / Hugging Face Spaces

The Dockerfile builds the Next.js app, installs Python, ffmpeg and openai-whisper, and serves on port 7860, which is the port the Space metadata above declares.

docker build -t podcast-transcribe .
docker run -p 7860:7860 podcast-transcribe

Stack

Next.js 16, React 19, TypeScript, Tailwind CSS 4, shadcn/ui components, Python with openai-whisper.

Credits

The audio player started from this tutorial; see Explanation_Personnal.md for the project notes.