Year: 2021
Type: conference-paper
Abstract: To quickly and accurately measure psychological well-being has been a challenging task. Traditionally, this is done with self-report surveys, which can be time-consuming and burdensome. In this work, we demonstrate the use of short voice recordings on smartphones to automatically predict well-being. In a 5-day study, 35 participants used their smartphones to make short voice recordings of what they were doing throughout the day. Using these recordings, our model can predict the participants’ well-being scores with a mean absolute error of 14%, relative to the self-reported well-being (“ground truth”). Both audio and text features from the recordings, especially, MFCC and semantic features, are important for prediction accuracy. Based on the work, we provide suggestions for future research to further improve the prediction result.
Authors: Yu-Ning Huang, Siyan Zhao, Michael L Rivera, Jason I. Hong, Robert E Kraut
Institution: Carnegie Mellon University
Language: English
Topic: Emotion and Mood Recognition
Subfield: Experimental and Cognitive Psychology
Field: Psychology
Domain: Social Sciences
Open Access status: hybrid
OpenAlex ID: W3161916366