Frequency Range of Fourier Phase Space as a Candidate Code for Musical Pitch Structure: Evidence from Artificial Neural Network Models of Iranian Classical Music
Year: 2026
Type: preprint
Abstract: How is musical pitch structure represented across cultures? Artificial neural networks trained on Western tonal classification problems spontaneously develop connection weights correlated with discrete Fourier phase spaces of the 12-tone pitch-class system. We test whether this extends to a structurally distinct micro-tonal system: Iranian classical music, whose 17 pitch classes lie on a 24-tone quarter-tone lattice and are not a subset or transposition of the Western chromatic scale. We defined nine classification tasks from Iranian music theory, covering melodic intervals and Dangs, trained Gaussian value-unit networks on each, and within the same pipeline trained matched networks on Western 12-tone tasks. Across every task, hidden units were reliably dominated by a single Fourier phase space: this exceeded a permutation null under false-discovery-rate control and outperformed pitch-height and random baselines. Iranian networks recruited the full range their lattice permits (1–12) and Western networks theirs (1–6). Excluding the Nyquist phase space, which on a quarter-tone lattice degenerates into a chromatic/microtonal partition, and matching task type, 31% of Iranian units occupy frequencies 7–11 that a 12-tone lattice cannot represent, against none of the Western units. A label-permutation control collapsed all fits to chance, indicating the structure is task-driven. Fourier phase space is thus a representational format networks discover beyond Western tonal music.
Source: Research Square
Authors: Yousef Aftabi, Hamid Farvaresh
Institution: University of Kurdistan
Language: English
Topic: Music and Audio Processing
Subfield: Signal Processing
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7225900462
