Predicting Emotional Valence of People Living with the Human Immunodeficiency Virus Using Daily Voice Clips: A Preliminary Study

• Main Authors: Ray F. Lin, Shu-Hsing Cheng, Yung-Ping Liu, Cheng-Pin Chen, Yi-Jyun Wang, Shu-Ying Chang
• Format: Article
• Language: English
• Published: MDPI AG 2021-09-01
• Series: Healthcare
• Subjects: HIV; speech emotion recognition; feature selection; artificial intelligence; clinical diagnosis
• Online Access: https://www.mdpi.com/2227-9032/9/9/1148

To detect depression in people living with the human immunodeficiency virus (PLHIV), this preliminary study developed an artificial intelligence (AI) model aimed at discriminating the emotional valence of PLHIV. Sixteen PLHIV recruited from the Taoyuan General Hospital, Ministry of Health and Welfare, participated in this study from 2019 to 2020. A self-developed mobile application (app) was installed on sixteen participants’ mobile phones and recorded their daily voice clips and emotional valence values. After data preprocessing of the collected voice clips was conducted, an open-source software, openSMILE, was applied to extract 384 voice features. These features were then tested with statistical methods to screen critical modeling features. Several decision-tree models were built based on various data combinations to test the effectiveness of feature selection methods. The developed model performed very well for individuals who reported an adequate amount of data with widely distributed valence values. The effectiveness of feature selection methods, limitations of collected data, and future research were discussed.