Speaker
Description
Turner syndrome is a rare chromosomal condition affecting females, characterized by high phenotypic heterogeneity and frequent delays in diagnosis. In recent years, digital health approaches have highlighted the potential of non-invasive biomarkers to support earlier and more accessible screening strategies. Within this context, the BeNeXT project aims to develop scalable diagnostic and prognostic tools by integrating multimodal phenotypic information (such as facial, body, gait and voice data) with machine-learning techniques.This contribution focuses on the exploration of voice as a digital biomarker for Turner syndrome. Speech signals were collected using different microphone types and a smartphone, reflecting realistic, low-cost acquisition scenarios. Multiple phonatory tasks, ranging from highly controlled productions to more complex speech activities, were considered in order to capture complementary aspects of voice production. Acoustic features related to phonation stability, resonance, prosody and temporal organization were extracted, and statistical analysis together with data-driven models were employed to assess their descriptive and discriminative potential.Preliminary analyses suggest that voice contains systematic acoustic patterns that may be associated with Turner syndrome. Beyond classification performance, key open challenges include ensuring robustness across recording conditions and improving the interpretability of the observed acoustic differences.Overall, this work supports the relevance of voice analysis as a promising component within multimodal pipelines for rare-disease characterization. Ongoing and future work within BeNeXT will extend these analyses to larger samples, additional speech tasks and integrated models, with the goal of advancing reliable, explainable and clinically meaningful voice-based biomarkers.