Narrated

Turns any article URL into a short, two-host podcast-style audio clip — not flat text-to-speech, but a generated dialogue between two AI voices discussing the piece.

Tech Stack

• Backend: FastAPI (Python)
• Article extraction: trafilatura
• Script generation: Claude API (Anthropic) — condenses the article into a two-host dialogue
• Text-to-speech: ElevenLabs API — one voice per host
• Audio stitching: pydub
• Frontend: React (Vite)

Key Features

• Paste any article URL and get back a playable MP3 narration
• Two distinct AI voices (Alex and Sam) hold a short conversational discussion of the article instead of reading it verbatim
• Article text extraction handles arbitrary web pages, not just a fixed set of sites
• Clean error handling for unfetchable pages, short/invalid articles, and upstream API failures

How It Works

flowchart LR
    A[Article URL] --> B["Scrape article\n(trafilatura)"]
    B --> C["Generate 2-host dialogue\n(Claude API)"]
    C --> D["Synthesize each line\n(ElevenLabs TTS)"]
    D --> E["Stitch clips together\n(pydub)"]
    E --> F["Return audio\n(FastAPI response)"]
    F --> G[Play in browser]

Loading

1. You paste an article URL into the frontend.
2. The backend fetches and extracts the readable article text.
3. Claude condenses that text into a short two-host dialogue (JSON: speaker + line).
4. Each line is sent to ElevenLabs with a distinct voice per host.
5. The resulting clips are stitched into a single MP3 and streamed back to the browser.

Screenshots

Default state	Filled in

Setup

Backend

cd backend
python -m venv .venv && source .venv/bin/activate  # or .venv\Scripts\activate on Windows
pip install -r requirements.txt
cp ../.env.example ../.env  # fill in ANTHROPIC_API_KEY and ELEVENLABS_API_KEY
uvicorn app.main:app --reload

Frontend

cd frontend
npm install
npm run dev

The app runs on http://localhost:5173 and talks to the backend on http://localhost:8000.

Why I built this

Wanted a hands-on project using ElevenLabs' TTS API alongside an LLM for content generation, rather than a generic chatbot wrapper — a good demonstration of a full audio-generation pipeline end to end.