ARGIRA
ICAD 2026 · 31st International Conference on Auditory Display
Crossmodal Correspondences Emerging from Painting Sonification
Jose Ranero García · Argira Station, Barcelona
argira.eus
Malevich · silence
─────────────────→
Kandinsky · chaos
"A blind person can tell a Monet from a Kandinsky apart — by sound alone." — Argira Research Hypothesis
Introduction

Paintings are experienced visually. Argira asks whether their structure can also be experienced through sound.

Access to visual art remains a barrier for blind and low-vision people. Verbal descriptions give only partial access. Sonification offers an alternative that preserves structure.

0.923 r · repl. hue→Ds
0.852 R² variance
74 works · N
The Pipeline
1
hue_std — standard deviation of the hue channel (HSV). Maps to fundamental frequency and activates OSC2, increasing timbral complexity.
2
Dv — visual fractal dimension via box-counting. Modulates granular texture.
3
Tempo — weighted sum of three contributions (capped at 15 BPM each): chromatic variability (hue_std), compositional asymmetry (luminance variance across 9 zones) and fractal complexity (Dv). Base 70 BPM, range 40–115 BPM.
Spectral analysis

The output WAV spectrogram is analysed to extract the spectral fractal dimension (Ds) via box-counting on log-scaled STFT.

Hann 1024-seg · overlap 768 · FFT 2048 · 80th percentile · scales 4–64

Accessibility
Spatial audio — stereo position encodes the horizontal chromatic centre of mass (cx). Mean error <0.05.
Temporal magnifier — slows to 0.5× to perceive chromatic texture in detail. Correlation r≈0.89 preserved at any speed.
Sweep mode — the sound evolves in real time as a finger slides across the artwork.
Spatial tour — zone-by-zone description, clockwise from 12 o'clock.

Monet's Poppies → high pitch, complex timbre (802 Hz, dual oscillator)

Van Gogh's Starry Night → dense texture, low frequency (272 Hz, high granular modulation)

Malevich → simple, quiet tone

Malevich · silence
Kandinsky · chaos →
Tap anywhere — selects the nearest point (74 works)
Map axes

X — hue_std: chromatic variability. Left = monochromatic, Right = full spectrum.

Y — Ds: spectral fractal dimension of the generated sound. Up = more complex.

Each point's colour encodes its position on X: blue → cyan → green → amber.

Open Live Perceptual Field
0.887 r original N=45
0.923 r replicated N=74
0.852 R² replicated
Multiple regression (N=45)

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 = 0.8417 · p < 0.001

Correlation table
CorrelationrNType
hue_std → Ds0.88745Designed
hue_std → Ds (replic.)0.92374Replication
hue_std → Roughness0.68730Emergent
hue_std → Rough. (no outlier)0.79029Emergent
local_contrast → freq_hz0.94974Structural
edge_density → effort0.75974Structural
Pilot perceptual test
5/5
N=5 participants · ages 22–78
Level 0 — Discrimination: all distinguished between two sonified works.

Dark-palette paintings → spontaneously associated with dark colours. Warm paintings → with bright tones. Regardless of musical background.
Discussion

Argira captures painting structure that is mathematically measurable and perceptually meaningful, without explicit perceptual rules.

The Dv suppression effect reveals that visual fractal complexity contributes to the auditory one independently of colour — mirroring findings in crossmodal correspondences.

Post-Review Extensions
6.1 Anti-Circularity (N=30)

An independent pipeline (band-limited sawtooth, no fractal metrics) confirms that hue_std predicts emergent spectral roughness: r=0.687 (r=0.790 excl. Malevich). A physical consequence of harmonic superposition, not of the algorithm's design. Replicated across 4 waveform types.

6.2 Corpus expansion (N=74, 32 artists)

Core correlation r=0.923 (R²=0.852, permutation p<0.001, N=1000). Fully consistent with the original result (r=0.887, N=45). Not an artifact of the initial selection.

6.3 Extended model A = f(hue_std, sat_mean, grad_mean)

Two images with identical hue_std=0.500 but opposite saturation produce completely different acoustic profiles (6 vs 13 emergent harmonics). Harmonic superposition is the critical generative mechanism. Pure sine waves reduce r from 0.887 to 0.44.

6.4 Closing the loop — Structural preservation (N=74)

Layer A (verified): stereo position preserves chromatic centre of mass, error <0.05.
Layer B (strong): local_contrast→sonic_freq_hz r=0.949; edge_density→effort r=0.759 (p<10⁻¹⁴).
Layer C (limit): direct inversion not currently feasible — a design choice, not a theoretical impossibility.

6.5 VAAM — Visual-Acoustic Association Matrix (N=117)

Systematic matrix of associations between 8 visual descriptors and 7 acoustic dimensions across 117 paintings. For each pair we compute S = (|r|+|ρ|)/2 (association strength) and δ = |ρ|−|r| (non-linearity index).

4 main findings:

Two independent families — chromatic (hue_std, hue_entropy_bits) → frequency; structural (edge_density, luminance_contrast) → texture and timing.

hue_std and hue_entropy_bits are not redundant — they independently predict distinct acoustic targets.

fractal_D: an honest null result — minimal predictive power (S max ≈ 0.27). Reported explicitly.

Roughness concentrates non-linearity — interaction effects not captured by linear models.

Key distinction: structure imposed by the operator (S≈1.000 in hue_std→odd_bias and luminance_contrast→decay_s) vs emergent corpus properties. That distinction is the central methodological contribution.

6.6 Interactive museum implementation

Argira Perceptual Field (v11): a continuous, navigable perceptual field running in any browser. 74 works organised in a 2D space (chromatic variety × visual complexity) with a Gaussian kernel (σ=0.18, k=7). Designed from feedback by a blind user (Felix).

Future work
Expand the perceptual test to low-vision participants — formal ABX + statistical power analysis.
Comparison condition with random mappings.
Mobile app for in-situ sonification in museums.
Sonification installation and full accessible exhibition.
Live Perceptual Field

74 works arranged in a two-dimensional space. The sound evolves continuously with cursor or touch movement — no discrete categories, no abrupt transitions.

Open Perceptual Field ↗ argira.eus/argira-sonification/ ↗
Data & code

CC BY-NC 4.0 · Zenodo
10.5281/zenodo.18865019

Datasets, source code, statistical analyses and extended experiments publicly available.

Key references

[1] Ramachandran & Hubbard — Synaesthesia (2001)

[2] Iosifyan et al. — Paintings & Music (2022)

[3] Spence — Crossmodal correspondences (2011)

[7] Taylor et al. — Fractal analysis of Pollock (1999)

ICAD 2026 · Jose Ranero García · CC BY-NC 4.0
ES EN
View in Perceptual Field →