MusicoCrypt: A Music-Modulated Stream Cipher Using Deep Pitch Estimation and Number-Theoretic Dynamics

This paper presents MusicoCrypt, a context-bound music-modulated stream-encryption framework that incorporates audio-derived public side information into a pseudorandom-function-centered encryption workflow. A shared reference recording is processed using a pitch-tracking pipeline, and ordered note pairs are encoded as residues in to form a contextual carrier. We first analyze MusicoCrypt-Core, a raw affine construction in which HMAC-SHA256-derived table indices select Happy-number and Collatz-based coefficients for carrier modulation. Motivated by the observed statistical bias of this raw affine stream, we introduce MusicoCrypt-W, a bias-resistant variant in which the audio-derived carrier and affine residue are used as domain-separated inputs to a final HMAC-SHA256 extraction layer with unbiased rejection sampling over . A byte-oriented MusicoCrypt-WB interface is also defined for binary protocols and standardized bit-level testing. The evaluation includes long-sequence stream statistics, HMAC-only/audio-hash/carrier ablations, NIST SP 800-22 bit-level tests, fixed-budget extraction analysis, known-plaintext analysis, frame-resynchronization simulation, acoustic synchronization experiments under additive white Gaussian noise, and computational-cost measurements that separate offline carrier extraction from online keystream generation. Results on symbols show that the raw Core keystream remains detectably non-uniform, whereas MusicoCrypt-W is consistent with the expected uniform distribution in the reported symbol-domain chi-square test. The ablation study shows that HMAC-based extraction provides the final statistical behavior, while the audio context provides context binding and cross-context separation. Accordingly, MusicoCrypt is positioned for context-bound multimedia workflows where offline carrier extraction is acceptable, rather than as a replacement for lightweight real-time binary stream ciphers.