Neuorise: Design and Implementation of an AI-Driven Adaptive Music Therapy System

Neuorise is an intelligent music therapy platform for patients with mild to moderate depression. It forms a complete personalized therapy loop by collecting user status via questionnaires, generating personalized prompts using Gemini, creating therapeutic music with Suno, and incorporating user feedback. Its aim is to provide auxiliary support for professional treatment.

Music therapy has a long history as a non-pharmacological intervention, but traditional programs mostly use pre-set tracks and struggle to adapt to individual emotional states in real time. Neuorise is positioned as an auxiliary treatment tool for patients with mild to moderate depression. Its core design philosophy is to enable the system to gradually learn and adapt to individual music preferences and emotional regulation needs by continuously collecting user feedback.

Neuorise's workflow forms a complete loop:

1. User Status Collection: Collect multi-dimensional data via 5-7 questionnaires + physiological data (heart rate, respiratory rate);
2. Prompt Generation: Input collected data into Google Gemini to convert into professional music generation prompts;
3. Music Generation: Create therapeutic music matching the current state via the Suno API;
4. Feedback and Regeneration: Users provide feedback via ratings/skipping, and the system optimizes subsequent generations based on this feedback for continuous iteration.

Backend Tech Stack

• Data Layer: SQLite stores user/session/questionnaire data, PBKDF2 hashing ensures security;
• Server: Python provides APIs, protected routes are only accessible to logged-in users;
• AI Integration: Gemini (via openai-proxy.org/google) and Suno API integration.

Frontend Interaction

Includes modules like questionnaire page, generation page, feedback page, history page, etc.

Audio Playback Strategy

• Play Suno audio directly when ready;
• Use synthetic previews during generation;
• Support seamless switching and volume adjustment.

1. Closed-Loop Feedback Mechanism: Unlike one-time recommendations, it implements a continuous learning process of collection → generation → feedback → regeneration;
2. LLM as Music Guide: Use Gemini to convert abstract emotions into professional music prompts, bridging the gap between emotion and music;
3. Multi-Modal Input Fusion: Combine subjective questionnaires and objective physiological data to fully depict the user's state;
4. Adaptive Music Therapy: Each generation is tailored to the current state, replacing the pre-set music library model.

Current Limitations

• Dependence on external APIs (Gemini/Suno) affects stability and cost;
• Suno generation delay may reduce real-time performance;
• Lack of large-scale clinical trials to verify effectiveness;
• Only an auxiliary tool, cannot replace professional medical care.

Future Improvements

• Integrate more alternative music generation models;
• Develop local lightweight generation capabilities;
• Collaborate with medical institutions for clinical validation;
• Expand to other mental health scenarios.

Neuorise combines the semantic understanding of large language models with AI music generation technology to create an intelligent system that understands emotions, generates personalized therapeutic music, and continuously learns. Although it is in the early stage, its design ideas such as closed-loop feedback and LLM as a music guide provide valuable references for the development of digital mental health tools, and it is expected to become an important auxiliary treatment tool in the future.