Acoustic Analysis of Speech and Voice in Parkinson’s Disease

Parkinson’s disease (PD) is the second most prevalent neurodegenerative disease following Alzheimer’s disease. Individuals with PD experience both motor and nonmotor symptoms secondary to decreased level of dopaminergic neurons in the substantia nigra, a substructure of the basal ganglia. Current medical and pharmaceutical intervention for PD consists of symptom mitigation and management. Most individuals with PD develop hypokinetic dysarthria, a motor speech disorder. Hypokinetic dysarthria is classified by reduced speech intelligibility, reduced loudness, monopitch, monoloudness, short rushes of speech and a breathy, harsh vocal quality. These speech and voice symptoms contribute to reduced speech intelligibility. Some changes in voice have been observed to arise during the prodromal disease stage, prior to an individual receiving a formal PD diagnosis. Acoustic measures provide an economic, noninvasive means to analyze changes in communication that occur secondary to PD, resulting in the potential to identify and track PD through an individual’s voice. The purpose of this thesis was to examine acoustic and perceptual analysis of both speech intelligibility and voice quality in speakers with PD. Both forms of analysis were further examined in relation to stage of Parkinson’s disease.

Data were collected from three different speech corpora: the Parkinson’s Disease Speech Corpus of Tampere University (PDSTU), the Healthy Adults Speech Corpus of Tampere University (HASTU) and the Parkinson’s Disease Speech Corpus of Tampere University long (PDSTUlong). Speech samples (continuous speech and vowel prolongation) from speakers with PD (n = 34), healthy controls (n = 15) and speakers with early-stage PD (n = 12) were gathered from the speech corpora. Speaker background information (sex, age, Hoehn & Yahr scores, Unified Parkinson’s Disease Rating Scale scores) was also gathered. Measures of speech intelligibility (automatic vowel articulation index (aVAI)), voice quality (Acoustic Voice Quality Index (AVQI)), and breathiness (Acoustic Breathiness Index (ABI)) were analyzed. Two perceptual measures of analysis were also used: direct magnitude estimation (DME) intelligibility ratings and the GRBASI (grade, roughness, breathiness, asthenia, strain, instability) scale were utilized to measure the efficacy of aVAI and AVQI respectively. Nonparametric statistical analysis was used to analyze the relationship between the above-mentioned acoustic measures, perceptual measures, and measures of PD stage.

The current gold standard for speech intelligibility analysis in the clinical setting is orthographic transcription. The current gold standard for voice quality ratings in the clinical setting is perceptual analysis. Though these methods are efficient, they are subject to human bias. Results of this thesis demonstrate the efficacy of the selected objective acoustic measures (aVAI, AVQI, ABI) in analyzing speech and voice from individuals with PD. aVAI has the potential to serve as a tool for automatic speech intelligibility analysis given the strong correlation found between aVAI and perceptual ratings of speech intelligibility among speakers with PD (Study I). AVQI was found to be a valid measure for use with people with PD and the harmonics-to-noise ratio was found to be an important single measure in analyzing voice quality from speakers from this population (Study II). ABI was found to be more sensitive than AVQI in the analysis of voice changes in speakers with early-stage PD (Study III). Finally, voice changes were observed in speakers with early-stage PD independent from changes in disease severity (Study III).