Automatic Sensing of Speech Activity and Correlation with Mood Changes

he association between social relationships and psychological health has been established fairly recently, in the last 30-40 years, relying on survey-based methods to record past activities and the psychological responses in individuals. However, using the self-reporting methods for capturing social behavior exhibits a number of shortcomings including recall bias, memory dependence, and a high end user effort for a continuous long-term monitoring. In contrast, automated sensing techniques for monitoring social activity, and in general, human behavior, has a potential to provide more objective measurements thus to overcome the shortcomings of self-reporting methods. In this paper, we present a privacy preserving approach to detect one component of social interactions - the speech activity, through the use of off-the-shelf accelerometers. Furthermore, we used the accelerometer based speech detection method to investigate the correlation between the amount of speech (which is an aspect that reflects the participation in verbal social interactions) and mood changes. Our pilot study suggested that verbal interactions are an important factor that has an impact on individuals’ mood, while the study also demonstrated the potential of automated capturing social activity comparable to the use of gold standard surveys