Characterizing the distribution of silent intervals in the connected speech of individuals with Parkinson disease

The purpose of the current investigation was to examine a statistical approach to differentiating shorter silent intervals (e.g., stop gaps) from longer silent intervals (e.g., pause) at varying syntactic locations within a reading passage to better characterize articulatory and prosodic aspects of speech timing in Parkinson disease (PD). Silent intervals 15 ms and longer were extracted from reading sample. Logarithmic transformation of the silent interval durations yielded a bimodal distribution. Gaussian Mixture Model analysis was used to statistically differentiate the first mode (Mode 1) that corresponded to short silent intervals from the second mode (Mode 2) that corresponded to longer silent intervals. The syntactic context surrounding each silent interval was also categorized. Results revealed that the large majority of silent intervals that occurred within a clause, phrase, or word were assigned to Mode 1, while the majority of silent intervals that coincided with sentence-ending punctuation were assigned to Mode 2. Results revealed that Mode 1 intervals were slightly, but significantly longer for speakers with PD (Mean = 52.48 ms, SE = 3.23) compared to controls (Mean = 44.67 ms, SE = 2.00). Examination of the surrounding syntactic context revealed that this difference occurred at between-word boundaries contained within a phrase or clause. No between group differences were observed for the other inter- and intra-sentence syntactic boundaries or Mode 2 intervals. This study outlines a data-based approach to differentiating short between- and within-word intervals from longer silent intervals or pauses that reflect the prosodic and syntactic structure of a reading passage. Using this approach, the current data suggest that speakers with PD exhibit longer short silent intervals than controls, potentially reflecting a slight delay in the fluent segment-to-segment transition between words.