Argira v3.5 — Multiple Regression Output: hue_std, Dv, Ds (N=45 paintings)Contributions to Spectral Fractality in the Sonification Pipeline

Argira v3.5 extends the statistical analysis of v3.4 to a larger and more diverse dataset of 45 paintings, validating the robustness of the relationship between visual features and spectral fractal dimension (Ds). Pipeline stability: maintains the optimized STFT configuration from v3.3 (Hann window, 1024-segment, 768-overlap, 2048-FFT, log1p scaling, adaptive binarization at 80th percentile, fixed box-counting scales 4–64). Dataset: 45 paintings from different authors and periods including Monet, Van Gogh (sunflowers series), Mona Lisa, and others.Simple correlations:hue_std vs Ds: r = 0.8869Dv vs Ds: r = -0.0381 (suppression effect confirmed)tempo vs Ds: r = 0.4210Partial correlation Ds–Dv controlling for hue_std: r = 0.4993Multiple regression (Ds ~ hue_std + Dv + tempo):β(hue_std) = 0.3265 (t = 12.981)β(Dv) = 0.1208 (t = 3.236)β(tempo) = -0.0002 (t = -0.721)R² = 0.8417R² with hue_std only = 0.7866Increase when adding Dv = +0.0552 Interpretation: chromatic variability (hue_std) remains the dominant predictor of spectral complexity across a larger and more diverse dataset (r=0.8869, R²=0.8417). Visual fractal dimension (Dv) contributes significantly once color is controlled. Tempo does not contribute meaningfully. The model shows strong robustness when scaling from 20 to 45 paintings. These findings reinforce the theoretical framework of Argira: the pipeline captures emergent crossmodal correspondences between visual and auditory complexity without explicit perceptual rules.