Argira Sonification Pipeline v3.4 — Dataset Report: Calibración empírica, validación arquitectural y comparativa v3.3 → v3.4

Argira is a painting sonification pipeline that maps visual features of artworks to acoustic parameters, generating a perceptual sound signature for each image. Argira Sonification Pipeline v3.4 reports empirical calibration results over a corpus of 187 images (43 canonical paintings, 21 Series I synthetic images, 123 batch synthetic images). The central result is that a three-channel orthogonal architecture where hue_entropy controls the base frequency (Channel A) and edge_density + fractal_D control modulation (Channel B) structurally reproduces the predictor inversion documented in ARGIRA V (10.5281/zenodo.20534328): the dominant predictor of visual roughness is not the dominant predictor of acoustic modulation.Channel B formula derived via OLS regression on N=43 canonical paintings (R² = 0.640):mod_depth = 144.9 + edge_norm×56.4 + fractal_norm×83.1 Key findings:• edge_density → mod_depth correlation improves from r=+0.694 (v3.3) to r=+0.900 (v3.4) — central validation of the OLS formula. • A previously undocumented phenomenon is identified: works with very different edge_density can produce near-identical mod_depth when fractal_D is equivalent. Van Gogh (Korenveld, edge=0.688) and Monet (Cliff Walk, edge=0.465) converge at mod_depth ≈ 244 Hz because their fractal_D is practically identical (1.904 vs 1.900). This behavior was not observable without residual analysis.• fractal_D is the dominant independent predictor of mod_depth once edge_density is controlled (r_partial = +0.507, p < 0.001), contributing ΔR² = +0.158 over edge_density alone. • roughness shows no independent signal in canonical paintings (r_partial = −0.128, p = 0.70) and is removed. Its apparent correlation in v3.3 (r=+0.249) was explained by collinearity with edge_density (r=+0.491), not by independent information. Changes from v3.3: OLS-derived Channel B formula, roughness removed, log-calibrated freq_base scale (150–750 Hz), column wav added, column granular removed. Files included:• sonify_painting_v3_4.py — full pipeline, standalone• sonificacion_v34.csv — complete dataset, 187 images, 28 variables• comparativa_v33_v34.csv — parametric comparison v3.3 → v3.4, 12 reference works• ARGIRA_v3_4_dataset_report.md — full report with analysis and comparisons• sonify_painting_v3_3.py / v3_2.py / v3_1.py — previous versions (reference)• sonificacion_v33.csv — v3.3 reference dataset fractal_vs_edge_by_artist.png — scatter fractal_D vs edge_density, 40 canonical paintings coloured by artist • Scatter analysis (fractal_vs_edge_by_artist.png) reveals artist-level clustering in the fractal_D × edge_density space: Rembrandt shows stable fractal_D across works with variable edge_density; Van Gogh shows high variance in both dimensions; Kandinsky and Turner occupy the high fractal_D / low edge_density quadrant.