Neural network-based measure of voice quality in Parkinson's disease

Parkinson's Disease (PD) significantly affects speech and voice. The Parkinsonian voice is often described as breathy, rough, hoarse, tremulous, abnormally pitched, having reduced pitch range, and unusually quiet. Changes in voice quality result from altered neurological controls of the muscles in the respiratory and phonatory systems, impacting breath support and vocal fold vibration. This study investigates the effects of PD on voice quality (e.g., breathiness) in vowel production among native Spanish speakers. The degree of breathiness is estimated from posterior probabilities calculated by recurrent neural networks trained to recognize spread glottis phonological features in Gujarati, a language contrasting breathy and modal voicing in vowels and between voiced aspirated (breathy voice) and voiceless aspirated versus plain voiced stops in consonants. It is hypothesized that vowels produced by PD patients will exhibit a higher degree of breathiness than those produced by normal controls, with degrees of breathiness potentially varying as a function of disease progression.