Voice abnormalities are common in Parkinson’s disease (PD), but the extent to which language-robust acoustic markers capture PD dysphonia in real-world clinical recording conditions and whether they are confounded by sex, language background, or medication state remains uncertain. This study aims to quantify PD-controlled differences in sustained-vowel acoustics in a Russian cohort, evaluate sex and language effects (Russian, Russian–Tatar bilinguals, and exploratory Tatar subgroup), and assess the robustness to clinical covariates. Cross-sectional data from the BRAINPHONE project were analyzed (n = 201; PD = 109; controls = 92). Participants produced sustained/aː/vowels in routine clinics (≥16 kHz, 32-bit.,wav). Acoustic features included perturbation (jitter and shimmer), cepstral/noise measures (cepstral peak prominence; harmonic-to-noise ratio; glottal-to-noise excitation ratio), and pitch metrics. Group contrasts used the Mann–Whitney U test and false discovery rate (FDR). Robust models adjusted for age, sex, and language; prespecified interactions probed diagnosis × sex/language. Spearman correlations related acoustics to Movement Disorder Society-Unified Parkinson’s Disease Rating Scale III, Hoehn and Yahr, disease duration, and medication variables. PD showed higher perturbation and lower cepstral/noise measures than controls (all q≤0.01 effects were consistent in females and males and replicated in Russian monolinguals and Russian–Tatar bilinguals with the Tatar monolingual subgroup being directionally similar. Covariate-adjusted models retained significant PD effects. Acoustic–clinical correlations were small (|ρ|≤~0.21) and did not survive FDR. In real-world clinical recordings of sustained vowels, CPPS, GNE, and shimmer provide robust, language-tolerant, medication-insensitive markers of PD dysphonia, supporting use as a complementary digital biomarker for telemedicine and longitudinal monitoring.