The automatic detection of pathological voices is challenging due to the variability and uncertainty of acoustic signals. This study addresses the problem through a dynamic representation that reflects the temporal variations of the electroglotography (EGG) signal’s patterns; exploring and comparing different clustering methods (classical, fuzzy, and intuitionistic) to generate recurrence plots that model this dynamics over time. These visual representations are used as input for the classification of healthy and pathological voices using various deep learning models. The results obtained with the Saarbrücken Voice Database (SVD) show that the intuitionistic fuzzy clustering (IFCM) leads to more expressive representations of temporal patterns, outperforming classical and fuzzy approaches. The Inception-v3 and ResNet-50 models achieve accuracies of up to 87% and 85%, respectively. Notably, IFCM offers practical advantages over more complex methodologies for the classification of vocal pathologies. These findings support the value of combining dynamic signal representation with comparative clustering strategies for improving pathological voice classification in clinical applications.