This study presents at NVIDIA GTC 2025 is a generative AI-based sleep music system designed to create soothing music tailored for different individuals through physiological feedback, aimed at improving sleep quality. The system employs GPT-SoVITSfor voice cloning, Musicgen, and Sunofor music generation, combined with a Retrieval-Augmented Generation (RAG) model for music similarity selection. Additionally, an embedding approach is used to convert audio files into vector data, facilitating similarity searches. Using physiological data from babies, we identify the most favorable music for them, retrieve similar audio files, and use Suno or Musicgento generate more suitable tracks, creating personalized music albums. The study included an initial test over 10 nights of sleep data and a user satisfaction survey. esultsindicated high satisfaction, with 85% of participants affirming that the system significantly improved sleep quality, and 90% reporting reduced stress. The sleep improvement: Implementation of the system resulted in a significant reduction in sleep onset time, averaging a 30% decrease.
The night awakenings decreased by 30%, contributing to better overall sleep quality. This study further explores how generative music can be adjusted through closed–loop adaptive human feedback to find music suitable for different situations.