Short-Term Detection of Mood Disorder Using Latent Affective Structure Modeling of Speech
碩士 === 國立成功大學 === 資訊工程學系 === 103 === Mood disorders, including unipolar depression (UD) and bipolar disorder (BD), are reported to be the most common mental illness in recent years. In diagnostic evaluation on the outpatients with mood disorder, a high percentage of BD patients are initially misdiag...
Main Authors: | , |
---|---|
Other Authors: | |
Format: | Others |
Language: | en_US |
Published: |
2015
|
Online Access: | http://ndltd.ncl.edu.tw/handle/12584458755523969622 |
id |
ndltd-TW-103NCKU5392071 |
---|---|
record_format |
oai_dc |
spelling |
ndltd-TW-103NCKU53920712016-08-15T04:17:47Z http://ndltd.ncl.edu.tw/handle/12584458755523969622 Short-Term Detection of Mood Disorder Using Latent Affective Structure Modeling of Speech 應用語音隱含式情感結構於情感性疾患之短期偵測 Yu-TingKuo 郭育婷 碩士 國立成功大學 資訊工程學系 103 Mood disorders, including unipolar depression (UD) and bipolar disorder (BD), are reported to be the most common mental illness in recent years. In diagnostic evaluation on the outpatients with mood disorder, a high percentage of BD patients are initially misdiagnosed as having UD. This results in significant negative consequences for the treatment of the BD patients. Therefore, it is crucial to establish an accurate distinction between BD and UD in order to make an accurate and early diagnosis, leading to improvements in treatment and course of illness. Given that speech is the most natural way for emotion expression, recognition of emotions in speech could be effectively applied to mood disorder detection. As current research focused on long-term monitoring of the mood disorders, short-term detection which could be used in early detection and intervention and thus reduce the severity of symptoms is desirable. This thesis proposes an approach to short-term detection of mood disorder based on the elicited speech responses. At first, eliciting emotional videos were used to elicit the patients’ emotions. Speech responses of the patients were collected through the interviews by a clinician after watching each of six emotional video clips. The support vector machine (SVM)-based classifier was adopted to obtain emotion profiles for each speech responses. In order to deal with the data bias problem, hierarchical spectral clustering algorithm were employed to adapt the eNTERFACE emotion database to fit the collected mood disorder database. The adapted eNTERFACE emotion data were then fed to the trained autoencoder to reconstruct the eNTERFACE emotion data for SVM-based emotion classifier construction. Finally, based on the emotion profiles generated from the SVM-based emotion classifier, a latent affective structure model (LASM) is proposed to characterize the structural relationship among the speech responses to six emotional videos for mood disorder detection. For system performance evaluation, speech responses were collected from 24 subjects, including 8 UD, 8 BD and 8 healthy people (control group) to construct the CHI-MEI mood database. Eight-fold cross validation was adopted for the following evaluation. Performance evaluation on the LASM-based approaches using autoencoder with different numbers of neurons and layers were conducted. The experimental results show that the proposed LASM-based method achieved 67%, improving by 9% accuracy compared with the commonly used classifiers like SVM and DNN. In future work, it will be helpful to improve system performance by integrating the proposed method with lexical and visual information. Furthermore, the individuality of the patient is also an important factor to be considered in mood disorder detection. Chung-Hsien Wu 吳宗憲 2015 學位論文 ; thesis 64 en_US |
collection |
NDLTD |
language |
en_US |
format |
Others
|
sources |
NDLTD |
description |
碩士 === 國立成功大學 === 資訊工程學系 === 103 === Mood disorders, including unipolar depression (UD) and bipolar disorder (BD), are reported to be the most common mental illness in recent years. In diagnostic evaluation on the outpatients with mood disorder, a high percentage of BD patients are initially misdiagnosed as having UD. This results in significant negative consequences for the treatment of the BD patients. Therefore, it is crucial to establish an accurate distinction between BD and UD in order to make an accurate and early diagnosis, leading to improvements in treatment and course of illness. Given that speech is the most natural way for emotion expression, recognition of emotions in speech could be effectively applied to mood disorder detection. As current research focused on long-term monitoring of the mood disorders, short-term detection which could be used in early detection and intervention and thus reduce the severity of symptoms is desirable.
This thesis proposes an approach to short-term detection of mood disorder based on the elicited speech responses. At first, eliciting emotional videos were used to elicit the patients’ emotions. Speech responses of the patients were collected through the interviews by a clinician after watching each of six emotional video clips. The support vector machine (SVM)-based classifier was adopted to obtain emotion profiles for each speech responses. In order to deal with the data bias problem, hierarchical spectral clustering algorithm were employed to adapt the eNTERFACE emotion database to fit the collected mood disorder database. The adapted eNTERFACE emotion data were then fed to the trained autoencoder to reconstruct the eNTERFACE emotion data for SVM-based emotion classifier construction. Finally, based on the emotion profiles generated from the SVM-based emotion classifier, a latent affective structure model (LASM) is proposed to characterize the structural relationship among the speech responses to six emotional videos for mood disorder detection.
For system performance evaluation, speech responses were collected from 24 subjects, including 8 UD, 8 BD and 8 healthy people (control group) to construct the CHI-MEI mood database. Eight-fold cross validation was adopted for the following evaluation. Performance evaluation on the LASM-based approaches using autoencoder with different numbers of neurons and layers were conducted. The experimental results show that the proposed LASM-based method achieved 67%, improving by 9% accuracy compared with the commonly used classifiers like SVM and DNN. In future work, it will be helpful to improve system performance by integrating the proposed method with lexical and visual information. Furthermore, the individuality of the patient is also an important factor to be considered in mood disorder detection.
|
author2 |
Chung-Hsien Wu |
author_facet |
Chung-Hsien Wu Yu-TingKuo 郭育婷 |
author |
Yu-TingKuo 郭育婷 |
spellingShingle |
Yu-TingKuo 郭育婷 Short-Term Detection of Mood Disorder Using Latent Affective Structure Modeling of Speech |
author_sort |
Yu-TingKuo |
title |
Short-Term Detection of Mood Disorder Using Latent Affective Structure Modeling of Speech |
title_short |
Short-Term Detection of Mood Disorder Using Latent Affective Structure Modeling of Speech |
title_full |
Short-Term Detection of Mood Disorder Using Latent Affective Structure Modeling of Speech |
title_fullStr |
Short-Term Detection of Mood Disorder Using Latent Affective Structure Modeling of Speech |
title_full_unstemmed |
Short-Term Detection of Mood Disorder Using Latent Affective Structure Modeling of Speech |
title_sort |
short-term detection of mood disorder using latent affective structure modeling of speech |
publishDate |
2015 |
url |
http://ndltd.ncl.edu.tw/handle/12584458755523969622 |
work_keys_str_mv |
AT yutingkuo shorttermdetectionofmooddisorderusinglatentaffectivestructuremodelingofspeech AT guōyùtíng shorttermdetectionofmooddisorderusinglatentaffectivestructuremodelingofspeech AT yutingkuo yīngyòngyǔyīnyǐnhánshìqínggǎnjiégòuyúqínggǎnxìngjíhuànzhīduǎnqīzhēncè AT guōyùtíng yīngyòngyǔyīnyǐnhánshìqínggǎnjiégòuyúqínggǎnxìngjíhuànzhīduǎnqīzhēncè |
_version_ |
1718376961555824640 |