A Conceptual AI-Assisted Music Creation Framework for Individuals with Amusia: Neural Pitch Correction and Adaptive Auditory Feedback

Amusia, a neurodevelopmental or acquired deficit in musical pitch perception affecting an estimated 2–4% of the population, excludes affected individuals from conventional music education and therapy because both rely on intact pitch self-monitoring. This paper addresses that gap by proposing the AI-Assisted Music Creation Framework (AAMCF), a conceptual five-layer architecture combining neural pitch estimation, key-snappingcorrection, transformer-based generative harmonisation, voice-cloning resynthesis, and multimodal adaptive auditory feedback. We formalise the amusia pitch-correction problem in information-theoretic terms, define theexpected-deviation bound that distinguishes amusic from neurotypical pitch perception, and introduce a closed-loop residual-correction model in which a lightweight neural network learns user-specific systematic pitch biases via online Bayesian optimisation. The proposed pipeline integrates established components — CREPE-basedpitch estimation, MusicGen-style harmonisation, and HiFi-GAN/VITS-based vocoding — into a single assistive architecture rather than proposing new low-level models. We present a structured qualitative comparison of the AAMCF against traditional music rehabilitation and music therapy across seven criteria, define four candidate evaluation metrics (mean absolute pitch error, pitch-correction acceptance rate, user engagement score, andfronto-temporal connectivity index) intended to guide future empirical validation, and examine ethical risks including dataset bias toward Western tonal systems, hallucinated harmonics, and biometric privacy. As no implementation or user study has yet been conducted, this work is explicitly positioned as a conceptual architecture paper; we specify the minimal empirical pilot required before clinical or commercial claims can be supported. The contribution lies in providing a reproducible, falsifiable design specification at the intersection of assistive technology, cognitive neuroscience, and generative AI.