Mood Disorder Detection from Speech Using LSTM-Based Emotion Profile Tracking and Mood Verification

論文翻譯標題: 基於長短期記憶模型於語音情緒剖面追蹤和病狀驗證進行情感性疾患之偵測
  • 周 佳惠

學生論文: Master's Thesis


In mental health disorder Unipolar Depression (UD) and Bipolar Disorder (BD) have become the most common mental illness A large portion of the BD patients is misdiagnosed as UD on initial presentation As speech is the most natural way to express emotion this thesis focus on tracking emotion profile of speech to build a short-term mood disorder detector for diagnosis assistance This thesis proposes an approach to short-term detection of mood disorder based on the elicited speech responses At first eliciting emotional videos are used to elicit the patients’ emotions Speech responses of the patients are collected through the interviews by a clinician after watching each of six emotional video clips As Deep Scattering Spectrum (DSS) can obtain more detailed energy distributions in frequency domain than the Low Level Descriptors (LLDs) this study combines LLDs and DSS as the speech features A domain adaptation method combining hierarchical spectral clustering (HSC) algorithm and denoising autoencoder is proposed to adapt the emotion database to the mood disorder database to alleviate the data bias problem in the emotion space The autoencoders is then adopted to extract the bottleneck features for dimensionality reduction Hidden Markov model (HMM) is applied to characterize the trajectory of emotion profiles Finally HMM-based verification is used to improve the detection performance of mood disorders This study collected the elicited emotional speech data from 15 BDs 15 UDs and 15 healthy controls Five-fold cross validation scheme was employed for evaluation Experimental results show that the proposed method achieved a detection accuracy of 73 33% improving by 18% compared to the SVM-based method In the future the patient’s personality response context and facial images can be considered for obtaining a better performance
獎項日期2016 8月 22
監督員Chung-Hsien Wu (Supervisor)