Assessing Bipolar Episodes using Speech Cues derived from Phone Calls

Abstract. In this work1 we show how phone call conversations can be used to objectively predict manic and depressive episodes of bipolar dis-ordered people. In particular, we use phone call statistics, parameters derived from dyadic phone conversations and emotional acoustic fea-tures to build and test user-specific classification models. Using random forest, we were able to detect the bipolar states with an average F1 score of 83 %, and we identified the speaking length and phone call length, the HNR value, the number of short turns and the variance of pitch F0 to be the most important variables for prediction