Comparing Thin-Slicing of Speech for Clinical Depression Detection

In recent years, speech based depression recognition has become a hot research spot. Previous researches shown that using smaller parts of the speech data will perform similarly if not better than using the whole speech data for depression recognition. In this study, 92 Chinese depressed patients and 92 age-, gender-and education level-matched control participants were examined to investigate the slicing theory in speech based depression recognition. By examining and comparing recognition results of different database of thin-slicing we found: 1) Small segments of continuous speech can still provide accurate clustering decisions, especially for females. When the segment is too short, the slicing effect becomes less obvious or even disappears. 2) The classification accuracy of speech segments is related to the location of the slice. The laws of them vary with the subjects' gender and speech patterns.