Turner syndrome (TS) is a rare chromosomal condition affecting only females and characterized by a wide phenotypic variability, which often delays diagnosis. This study examines the potential of vocal biomarkers to improve TS diagnosis using sustained vowel recordings from 62 Brazilian participants (43 controls, 19 with TS; ages 16-55) collected with a smartphone. Prosodic, spectral, and formant-related features were extracted using OpenSMILE and Praat. The statistical analysis revealed significant differences in several formant and shimmer descriptors. Four supervised machine learning models were trained to classify speech recordings between TS and control. Several models achieved fair discrimination for this binary task, with AUC values between 0.7 and 0.8. These findings motivate validation in larger cohorts and the inclusion of additional speech tasks, with the long-term goal of developing non-invasive low-cost voice analysis as a complementary tool for TS screening.