Objective: This study first evaluates the reliability of existing tools for the automatic analysis of speech prosody in Parkinson's disease. It then investigates what acoustic measurements can be used to augment perceptual descriptions of Parkinson's speech. Background: Analyzing (dys)prosody in neurological disorders is not an easy task. Preliminary studies on dysprosody were based on perceptual assessment but as it is generally accepted perceptual descriptions need to be completed with more objective ones. This requires the development of tools and measurements The Movement Disorder Society : Abstract # 551891 Preview http://www.call4abstracts.com/mds/finalpreview.php?absnu... 1 sur 2 25/06/12 12:11 suitable for the prosodic analysis of pathological speech. Methods: In this study, three algorithms were first evaluated for (1) the automatic measurement of pitch extrema, (2) syllable detection and (3) pause detection. Analyses based on global vs. interpausal runs measurements were then carried out on the read and semi-spontaneous speech of 25 control and 30 Parkinson subjects (off state). Results: Results show that the three algorithms are robust enough for the automatic analysis of Parkinson's speech, provided that measurements are carried out from inter-pausal runs and not from the whole recording. This study also reveals that Parkinson's speech (off state) is characterised by smaller variations in pitch range, voice intensity and articulation rate. These results suggest that the perceptual impression of monotonous voice in Parkinson' speech may not (only) be due to a pitch range narrowing but rather to an impairment in varying pitch, intensity and duration. Results also reveal that whereas control subjects increase variations in pitch range, voice intensity and articulation rate in semi-spontaneous speech, Parkinson's patients do not and still display small prosodic variations. This study therefore suggests that some linguistic functions may not be conveyed in read and in particular in semi-spontaneous Parkinson's speech. Conclusions: This study has proved that existing tools for the automatic analysis of speech prosody are suitable for the analysis of Parkinson speech, provided that prosodic characteristics are measured at different anchor points and that these measurements are made on large corpora. Analyses are currently being carried out to investigate the effect of L-Dopa on prosodic variations in Parkinsons' speech.