计算机科学
Mel倒谱
均方误差
萧条(经济学)
模态(人机交互)
人工神经网络
语音识别
人工智能
机器学习
模式识别(心理学)
特征提取
统计
数学
宏观经济学
经济
作者
Emna Rejaibi,Ali Komaty,Fabrice Mériaudeau,Said Agrebi,Alice Othmani
标识
DOI:10.48550/arxiv.1909.07208
摘要
Clinical depression or Major Depressive Disorder (MDD) is a common and\nserious medical illness. In this paper, a deep recurrent neural network-based\nframework is presented to detect depression and to predict its severity level\nfrom speech. Low-level and high-level audio features are extracted from audio\nrecordings to predict the 24 scores of the Patient Health Questionnaire and the\nbinary class of depression diagnosis. To overcome the problem of the small size\nof Speech Depression Recognition (SDR) datasets, expanding training labels and\ntransferred features are considered. The proposed approach outperforms the\nstate-of-art approaches on the DAIC-WOZ database with an overall accuracy of\n76.27% and a root mean square error of 0.4 in assessing depression, while a\nroot mean square error of 0.168 is achieved in predicting the depression\nseverity levels. The proposed framework has several advantages (fastness,\nnon-invasiveness, and non-intrusion), which makes it convenient for real-time\napplications. The performances of the proposed approach are evaluated under a\nmulti-modal and a multi-features experiments. MFCC based high-level features\nhold relevant information related to depression. Yet, adding visual action\nunits and different other acoustic features further boosts the classification\nresults by 20% and 10% to reach an accuracy of 95.6% and 86%, respectively.\nConsidering visual-facial modality needs to be carefully studied as it sparks\npatient privacy concerns while adding more acoustic features increases the\ncomputation time.\n
科研通智能强力驱动
Strongly Powered by AbleSci AI