Proteins are fundamental to life, with their function determined by a complex interplay of sequence, structure, and dynamics, yet traditional methods often struggle to capture these multifaceted relationships comprehensively, limiting our understanding and predictive capabilities. This study introduces an innovative interdisciplinary approach, translating protein sequence and structural information into musical encodings, to explore these multidimensional functional relationships through spectral analysis. By applying machine learning models to the musical spectra of over 2,000 proteins, we achieved high classification accuracy (91.04\%) and demonstrated a significant correlation between a proposed musical harmony index and protein functionality; furthermore, we successfully integrated this music encoding with diffusion models for the guided directed evolution of Green Fluorescent Protein (GFP), showcasing its utility in protein engineering. This research offers a novel analytical framework and practical tool for protein analysis, functional prediction, and design, providing new insights for understanding and engineering the "symphony of life."