Center of Clinical Neuroscience, Charles University in Prague
This presentation aims to guide you through the fundamentals of acoustic vocal biomarkers in motor speech disorders and demonstrate how it can be useful to phoneticians.
The data included 570 native Czech speakers, including 42 subjects with idiopathic rapid eye movement sleep behavior disorder (RBD), 32 with early, untreated Parkinson’s disease (PDU), 26 with treated Parkinson’s disease (PDT), 22 with multiple system atrophy (MSA), 15 with progressive supranuclear palsy (PSP), 18 with untreated Huntington’s disease (HDU), 13 with treated Huntington’s disease (HDT), 17 with cerebellar ataxia (CA), 101 with multiple sclerosis (MS), and 284 healthy controls (HC) with no neurological or communication disorders. Please see Hlavnička (2018a) for more details about the cohort. Each speaker performed tasks that included sustained vowels (/A/ and /I/), a rhythm test, reading, a monologue, and a diadochokinetic task. The recordings were made in a quiet, non-reverberant room using a headset condenser omnidirectional microphone.
I analyzed speech features related to prosody, articulation, resonance, phonation, and respiration calculated via the Dysarthria Analyzer software (https://www.dysan.cz/, Hlavnička 2018b). To allow comparisons, all measured values were normalized. A subgroup of healthy controls, matched for age and sex with the individuals in each clinical group, was used to calculate z-scores. These z-scores were compared across all groups using one-way ANOVA for normally distributed features, and Kruskal-Wallis tests for gamma-distributed features. The significance level was set at 0.05.
Additionally, I conducted a classification experiment to identify hypokinetic and hyperkinetic speech tendencies (characterized by reduced movement amplitude) and excitatory speech tendencies (characterized by exaggerated movements due to lack of coordination or involuntary muscle contractions).
Among healthy controls (HC), only a small number exhibited significant hypokinetic (4%), hyperkinetic (2%), or unspecific (2%) speech tendencies. Hypokinetic speech patterns were most common in PDU (56%), PDT (67%), MSA (64%), PSP (34%), and less so in RBD subjects (18%), who are at high risk for developing Parkinson’s disease (PD). Hypokinetic tendencies were also observed in MS (23%), CA (15%), HDT (10%), and HDU (18%), often alongside hyperkinetic patterns. Hyperkinetic speech tendencies were particularly frequent in MS (31%), CA (31%), and most notably in HDU (73%) and HDT (100%). A significant increase in hyperkinetic tendencies was found in MSA (28%) compared to PSP (7%, z-test, p < 0.01).
Tendencies to hypokinesia and hyperkinesia were found to be consistent with underlying hypotheses about disease manifestations. Observed clusters could have diagnostic value for recognizing Parkinson’s disease (PD) and atypical parkinsonian syndromes (APS). The results suggests that acoustic signs may be more distinct in early stages, before broader speech impairments occur due to speaker’s compensation strategies. Factors like patient history, oral exams, and perceptual findings should be considered individually to assess whether a feature serves as a disease marker or indicates progression.
Additionally, applicability of vocal biomarkers on other languages will be demonstrated on a large cohort including Czech, German, English, French, and Italian language collected within the multicentric international project (Rusz et al. 2021).
This analysis has been developed into a comprehensive methodology available at https://www.dysan.cz. A free e-mail course on how to start with vocal biomarkers of motor speech disorders is available at https://events.vocalgebra.com/first5speakers. Please feel free to contact the author at jan.hlavnicka@dysan.cz.