Listening app and story library
Working buildLantern Tales
Real moral stories, narrated aloud.
A listening app for moral stories, parables and true accounts of people who kept going. Each story comes with its lesson and a one-line takeaway, can be narrated in many accents, and is recommended based on what the listener loves.
A working application run locally. No public deployment has been confirmed.
Problem
Story collections are long lists. Finding the right story for how you feel right now is hard, and reading is not always possible.
Approach
Let people describe what they need in their own words, rank stories with full-text and semantic search together, learn taste from likes and listening, and narrate with neural voices, background soundscapes and a tone to suit each story.
Features
- Find a story in your own words
- Hybrid search: PostgreSQL full-text ranking fused with sentence-embedding similarity, plus length and tradition constraints pulled out of the request.
- Learns your taste
- Time-decayed taste profile, a per-listener logistic regression once there are enough likes and dislikes, and MMR diversification. Each pick explains why.
- Accented narration
- Free Microsoft Edge neural voices across many accents, audio cached per sentence, with device voices as an offline fallback.
- Narrate in your own voice
- Local voice cloning with Chatterbox after a spoken consent statement. Recordings stay on the machine and can be deleted.
- Synthesised soundscapes
- 19 original background sounds generated with numpy and scipy, so no recordings carry a licence.
- Auto-tagging
- Theme tags assigned zero-shot from each story's embedding; missing traditions inferred from nearest neighbours.
Next system
YatraMatrix