Generating AI-driven Interactive Bedtime Stories Based on a Child’s Mood and Preferences

Abstract

This paper presents an AI-powered interactive storytelling system that generates personalized bedtime stories tailored to a child’s emotional state and individual preferences. Traditional storytelling methods, while comforting, are static and fail to adapt to a child’s fluctuating moods, potentially limiting emotional engagement and therapeutic potential. Our system addresses this limitation by integrating emotion detection, narrative generation, and emotionally expressive speech synthesis into a unified framework. We employ a fine-tuned BERT-based model to identify the child’s emotional tone from text inputs, which then informs prompt construction for large language models—specifically, LLaMA fine-tuned with Low- Rank Adaptation (LoRA) or Gemini 1.5 Pro—to generate mood- aligned storylines. These narratives are subsequently converted into natural, emotionally resonant speech using cutting-edge TTS systems like XTTS or Eleven Labs. A web-based interface, built with React and Flask, allows caregivers to input mood cues and instantly access AI generated storytelling experiences in real time. Experimental results show a classification accuracy of 75.4