This work presents an automatic analysis of Parkinsonian speech prosody in read and spontaneous tasks. We first evaluate the reliability of existing algorithms for the automatic analysis of speech prosody and show that these tools are efficient for the analysis of Parkinsonian speech if a large amount of data is processed. We then investigate the prosodic characteristics of 25 control and 30 Parkinsonian subjects (in off (non-medicated) state) in both tasks and show that Parkinsonian read and spontaneous speech are characterized by longer pauses as well as smaller variations in pitch range, voice intensity level and articulation rate; these characteristics being all the more marked in semi-spontaneous speech. These results suggest that some linguistic functions, normally expressed via prosodic variations, may not be conveyed in Parkinsonian speech, particularly in a semi-spontaneous task.

Index Terms: prosodic variations, automatic measurements and tools, interpausal run, Parkinson disease, dysprosody