news-voice-app

News Voice UI

A web dashboard that groups today's top headlines into major story clusters and plays each as a 2-3 minute audio summary.

Features

• Polls BBC World, NPR, Guardian, and Al Jazeera every 30 minutes via RSS
• Groups articles into major events using GPT-4o-mini
• Generates a 2-3 minute broadcast script per group, stored as individual JSON files
• Only re-summarizes groups whose article set has changed since the last sync
• Prunes articles older than 24 hours automatically
• Card grid UI with images sourced from RSS feeds
• Click a card to play its audio summary; click again to stop; clicking another card switches
• Expand button opens a modal with the full summary text and source article links

Requirements

Base dependencies are installed via the root requirements.txt. See the main README for setup. Add SMALLEST_API_KEY and OPENAI_API_KEY to your .env.

Backend

cd backend
uv pip install -r requirements.txt

Frontend

cd frontend
npm install

Usage

# Terminal 1 — backend (runs on :8000, syncs news on startup)
cd backend && uvicorn main:app --reload --port 8000

# Terminal 2 — frontend (runs on :3000)
cd frontend && npm run dev

Open http://localhost:3000. Cards appear once the first sync completes (a few seconds after the backend starts).

How It Works

1. On startup the backend runs a sync immediately, then every 30 minutes via APScheduler.
2. Sync loads data/index.json, deletes article files older than 24 hours, then fetches fresh RSS from all four feeds.
3. New articles are written as individual files under data/articles/{id}.json and added to the index.
4. All current article titles are sent to GPT-4o-mini, which returns a JSON grouping of indices into major story clusters.
5. For each group whose article set has changed, GPT-4o-mini writes a 2-3 minute broadcast script saved to data/groups/{slug}.json.
6. GET /news reads the index and group files and returns the full card data to the frontend.
7. Clicking a card POSTs the stored summary text to POST /audio, which streams audio from Smallest AI TTS and returns a WAV file for immediate playback.

File Structure

news-voice-app/
├── .env.sample
├── backend/
│   ├── main.py          # FastAPI app + APScheduler + /news and /audio endpoints
│   ├── rss.py           # RSS fetching via feedparser
│   ├── ai.py            # OpenAI grouping and summarization
│   └── requirements.txt
├── frontend/
│   ├── package.json
│   └── app/
│       ├── layout.tsx
│       └── page.tsx     # Card grid, modal, and audio playback logic
└── data/                # Created at runtime, gitignored
    ├── index.json
    ├── articles/
    └── groups/

API Reference

• Waves TTS Overview
• Lightning v3.1 Stream endpoint

Next Steps

• News Voice App — CLI version of the same concept (Python only)
• Voice Explorer — Browse and preview all available voices