Timing patterns of speech as potential indicators of near-term suicidal risk
In an effort to find a reliable method that could assist clinicians in risk assessment, information in the speech signal has been found to contain characteristic changes associated with high risk suicidal states. This paper addresses the questions of (1) Does information contain in the speech timing...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IJMCR
2015
|
| Subjects: | |
| Online Access: | http://irep.iium.edu.my/47065/1/47065_-_Timing_patterns_of_speech_as_potential_indicators_of_near-term_suicidal_risk.pdf http://irep.iium.edu.my/47065/ http://ijmcr.com/timing-patterns-of-speech-as-potential-indicators-of-near-term-suicidal-risk/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | In an effort to find a reliable method that could assist clinicians in risk assessment, information in the speech signal has been found to contain characteristic changes associated with high risk suicidal states. This paper addresses the questions of (1) Does information contain in the speech timing-based measures able to discriminate between high risk suicidal (HR) speech from the depressed (DP) speech. (2) How well do speech features, specifically the timing-based measures can predict the ratings from a well-known medical diagnostic tool known as the Hamilton Depression Rating Scale (HAMD). In the first study, using the leave-one-out procedure as a means to measure a classifier performance for all-data classification revealed a single speech timing-based measure to be a significant discriminator with 79% overall correct leave-one-out classification in male (MR) and female (FR) reading speech from Database A. For male patients, using the trained features on Database A and testing on Database B1 successfully demonstrated up to 100% detection of high risk speech in Database B1. In the second study, the acoustic measurements were shown to effectively predict the HAMD score with less than 5% mean absolute error using only combinations from the timing-based measures and eliminating all spectrum-based measures |
|---|