Algorithmic Estimation of Pauses in Extended Speech Samples of Dysarthric and Typical Speech

The aim of this study was to evaluate the validity and performance of an algorithm designed to automatically extract pauses and speech timing information from connected speech samples. Speech samples were obtained from 10 people with amyotrophic lateral sclerosis (ALS) and 10 control speakers. Pauses were identified manually and algorithmically from digitally recorded recitations of a speech passage that was developed to improve the precision of pause boundary detection. The manual and algorithmic methods did not yield significantly different results. A stepwise analysis of three different pause detection parameters revealed that estimates of percent pause time were highly dependent on the values specified for the minimum acceptable pause duration and the minimum signal amplitude. Consistent with previous reports of dysarthric speech, pauses were significantly longer and more variable in speakers with ALS than in the control speakers. These results suggest that the algorithm provided an efficient and valid method for extracting pause and speech timing information from the optimally structured speech sample.