Cross-Modal Transformer Framework for Emotion-Aligned Music Therapy using Indian Classical Raagas for Individuals with Autism Spectrum Disorder

Autism Spectrum Disorder (ASD) is complicated and usually involves non-verbal, sensory sensitive and personalised therapeutic interventions. In this paper, we introduce a novel AI based framework, NeuroMusical Cross Modal Transformer (NM XMT) which translates self-expression of an autistic person in form of listening to song to an AI based recommendation system of emotionally and contextually contoured Indian classical Raagas for an autistic person. The system proposed here utilizes crossmodal deep leaning to embed clinical behavioral features and musical semantics into a common affective space, so that precise alignment of the therapy can be achieved using arousal valence modeling, affective sensory preference, and contextual parameters like the time of the day and age. Based on the simulated datasets, the model was evaluated against traditional and emotion aware baselines. The results indicate that NM-XMT performs superior to conventional recommender systems with a high therapeutic effectiveness score of 0.89, which is the best amongst compared models. Reinforcement-driven personalization and feedback loops mean that the system also has high explainability and adaptability. The significance of the model is that it can serve as the culturally grounded, non-invasive digital therapy solution for the ASD community, as the results from these findings show.