| Depression is a common mental disease,which is characterized as prominent and persistent low mood.Current diagnosis depends heavily on patients’ self-report and clinicians’ experience.Subjective biases during diagnosis lead to a high misdiagnosis rate.Therefore,an objective accurate convenient method for depression detection is necessary.Speech is an attractive candidate for its advantages: non-invasive,fast,convenient and economic.Speech slice refers to a small segment of speech obtained from continuous speech.This paper mainly focuses on whether the use of small segment of continuous speech can still provide accurate clustering decisions or not and several key problems in segment-based speech depression recognition.Studying on segment-based speech depression recognition can not only makes the study of speech based depression recognition more focused and reduces the workload from the source,but also helps improve the recognition accuracy.In this paper,we constructed two speech databases for slice theory researching and made a thorough research on segment-based speech depression recognition.The main contributions and innovations are as follows:1.Designed an experiment for depression speech data acquisition,which including interviews,words reading,text reading,picture description and TAT five paradigms,and positive,neutral and negative three emotional valence.Two speech databases for slice theory researching are constructed with 348 subjects.2.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.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.3)When the segment is too short,the slicing effect becomes less obvious or even disappears.3.Proposed a decision level fusion method based on slice location,which improves the recognition accuracy of depression.In this paper,we have made a comprehensive and in-depth study on segment-based speech depression recognition,verified the feasibility of depression recognition based on speech segments and analyzed several key problems of it,which has a certain reference value for the research in this field. |
