ARGIRA (IV): Structural Asymmetry in Sonification Mappings

ARGIRA (IV): STRUCTURAL ASYMMETRY IN SONIFICATION MAPPINGS

OVERVIEW

This repository contains the complete code, data, and results for ARGIRA IV, a study investigating how visual image descriptors relate to the outputs of two algorithmic sonification mappings: OPRS and RTR.

Previous phases of the ARGIRA research program (ARGIRA I–III) attempted to predict the divergence score:

Δ = OPRS − RTR

from visual image characteristics.

Across multiple representational approaches—including low-level image statistics, engineered feature spaces, semantic embeddings, and multimodal language-model descriptors—predictive performance remained near zero or negative under cross-validation.

ARGIRA IV revisits this problem by modeling OPRS and RTR independently rather than modeling their difference directly.

RESEARCH QUESTION

Do OPRS and RTR preserve different types of visual information from the same source image?

Instead of predicting the composite score Δ, this phase evaluates the visual predictability of each sonification mapping separately.

DATASET

Input dataset:

dataset_canonico_86_v1.csv

The analyzed subset includes valid image records for which all required visual predictors and sonification outputs were available.

Visual predictors:

• Naive Roughness
• Hue Standard Deviation
• Hue Entropy

Targets:

• OPRS Acoustic Roughness
• RTR Acoustic Roughness

METHODOLOGY

The experiment evaluates OPRS and RTR independently using:

• Ridge Regression
• Random Forest Regression
• 5-Fold Cross-Validation
• Permutation Testing

Performance is measured using cross-validated R².

Feature importance is estimated using Random Forest models.

MAIN RESULTS

OPRS

• Ridge Regression: R²CV = 0.582
• Random Forest: R²CV = 0.528

OPRS exhibits substantial predictability from visual image descriptors.

The dominant predictor is Naive Roughness, which accounts for most of the predictive signal.

RTR

• Ridge Regression: R²CV = 0.247
• Random Forest: R²CV = 0.086

RTR shows considerably weaker predictability.

Its behavior appears more strongly associated with chromatic descriptors than with texture-related measures.

INTERPRETATION

The results suggest that OPRS and RTR preserve different aspects of visual information.

Rather than indicating an absence of image–sound relationships, the negative findings reported in ARGIRA I–III may partly arise from modeling the difference score:

Δ = OPRS − RTR

as a single target variable.

When analyzed independently, the two mappings display distinct predictive structures, indicating that they transform visual information into sound through different mechanisms.

CENTRAL FINDING

Two sonification mappings applied to the same image can preserve radically different visual characteristics and generate distinct predictive structures in the image-to-sound transformation process.

This reveals a measurable structural asymmetry between sonification mappings.

REPOSITORY CONTENTS

• experimento_argira4_oprs_vs_rtr.py — complete analysis pipeline

• dataset_canonico_86_v1.csv — canonical ARGIRA dataset

• argira4_resultados.csv — model performance metrics

• argira4_figura.png — summary visualization of the principal findings

• README.md — reproducibility documentation

REPRODUCIBILITY

Environment:

• Python 3.10+
• NumPy
• Pandas
• Scikit-learn
• Matplotlib
• SciPy

Run:

python experimento_argira4_oprs_vs_rtr.py

The script reproduces:

• Cross-validation results
• Correlation analyses
• Permutation tests
• Feature importance estimates
• Publication figures

RELATED ARGIRA DEPOSITS

ARGIRA I

Visual Predictors of Acoustic Perceptual Distance in Image Sonification

DOI: 10.5281/zenodo.20524644

ARGIRA II

From Correlation to Failure: Limits of Visual Feature Spaces in Predicting Perceptual Sonification Differences

DOI: 10.5281/zenodo.20526610

ARGIRA III

Systematic Representational Failure — LLM Semantic Descriptors Cannot Predict Perceptual Sonification Divergence

DOI: 10.5281/zenodo.20530682

AUTHOR

Jose Ranero García

ARGIRA Research Project

2026