Novel robust decision support tool assisting early diagnosis of pathological voices using acoustic analysis of sustained vowels

Effective vocal communication is critical in daily life, and 30% of the general population may suffer from a voice disorder at some point in their lives. Early diagnosis of voice pathologies facilitates mitigating symptoms and optimizing treatment for expedient recovery. Here, we studied the potential of an automated clinical decision support tool to differentiate subjects with early onset voice disorders from healthy controls simply on the basis of a single sustained vowel phonation. We characterized 200 phonations from 200 subjects with 445 speech signal processing algorithms, extracting clinically useful properties of the phonations in order to differentiate healthy and pathological cases. We selected parsimonious gender-dependent feature subsets and demonstrated that we can automatically differentiate healthy and pathological subject cohorts with approximately 91% overall accuracy. These compelling findings endorse the use of the proposed methodology towards assisting speech experts in vocal performance assessment and diagnosis of early onset voice disorders.