ARGIRA VII: Visual–Acoustic Association Matrix (VAAM) for a Canonical Painting Corpus (N=117)

This dataset accompanies ARGIRA VII, a study within the ARGIRA (Aesthetic Research on Generative Image-to-Audio Relations) series. The objective is to characterize how visual image properties are reflected in acoustic parameters under a fixed image-to-sound mapping operator. We introduce the Visual–Acoustic Association Matrix (VAAM), a systematic representation of the association structure between visual descriptors extracted from paintings and the acoustic parameters generated through sonification. The analysis is performed on a canonical corpus of 117 paintings using a single sonification operator (M1). For each pair (visual variable, acoustic dimension), two complementary measures are computed: S(Vᵢ, Aⱼ) = (|r| + |ρ|) / 2 where r is the Pearson correlation coefficient and ρ is the Spearman rank correlation coefficient. S represents the overall strength of association by combining linear and monotonic structure. δ(Vᵢ, Aⱼ) = |ρ| − |r| a nonlinearity index indicating the extent to which monotonic relationships exceed what is captured by linear correlation alone. Visual Variables The corpus includes five primary visual descriptors: hue_std hue_entropy_bits edge_density fractal_D luminance_contrast Three interaction terms are additionally evaluated: edge_x_fractal hue_x_edge hue_x_fractal Acoustic Dimensions The corresponding acoustic dimensions are: freq_base_hz n_harmonics odd_bias mod_rate_hz roughness tempo_bpm decay_s Main Findings Four principal findings emerge from the analysis. 1. Chromatic and Structural Families Visual predictors organize into two functionally independent families. A chromatic family (hue_std and hue_entropy_bits) preferentially maps onto frequency-related acoustic dimensions, whereas a structural family (edge_density and luminance_contrast) preferentially maps onto temporal and texture-related dimensions. 2. Independence of Hue Descriptors hue_std and hue_entropy_bits are demonstrated to be non-redundant observables. Partial-correlation analyses show that they independently predict different acoustic targets. In particular, hue_entropy_bits remains strongly associated with frequency-related dimensions after controlling for hue_std, while hue_std uniquely explains odd-bias structure. 3. Acoustic Inertia of Fractal Dimension fractal_D exhibits minimal predictive power across all acoustic dimensions (maximum association approximately S = 0.27). This result contrasts with the documented relevance of fractal measures in empirical aesthetics and suggests that fractal information may either be poorly transmitted by this class of sonification operator or require alternative mapping architectures to become acoustically accessible. 4. Nonlinearity Concentrated in Roughness The acoustic dimension roughness concentrates the largest nonlinearity indices in the matrix. Multiple predictors show substantially stronger monotonic than linear relationships with roughness, indicating the presence of interaction effects and nonlinear structure not captured by additive linear models. Operator-Imposed vs Emergent Structure Two associations reach near-perfect values (S ≈ 1.000): hue_std → odd_bias luminance_contrast → decay_s These relationships are not interpreted as empirical discoveries. They arise directly from deterministic equations implemented in the sonification operator and are therefore reported explicitly as operator-imposed structure rather than emergent corpus properties. Distinguishing imposed relationships from observed associations is a central methodological contribution of this dataset. Scope and Limitations The VAAM should be interpreted as a measurement of association under a specific operator rather than evidence of cross-operator invariance. The present work establishes the empirical association layer of the ARGIRA framework. The question of invariant survival—whether visual properties preserve their acoustic associations across structurally different sonification operators—remains open and constitutes the primary objective of ARGIRA VIII. Contents File Description sonificacion_resultados.csv Complete dataset (117 canonical paintings and 110 synthetic controls) argira_vii_invariants.py Reproducible analysis script argira_vii_figures.py Figure-generation script argira_vii_heatmap.png Visual–Acoustic Association Matrix argira_vii_survival_profiles.png Association profiles by visual variable argira_vii_hue_question.png Comparative analysis of hue-based descriptors argira_vii_Sdelta.png Combined S and δ matrices argira_vii_clustering.png Hierarchical clustering of association structure README.md Project documentation Citation Ranero García, J. (2025). ARGIRA VII: Visual–Acoustic Association Matrix for a Canonical Painting Corpus (N=117). Zenodo. DOI: 10.5281/zenodo.20554745.