Diagnosing Bipolar Disorders in a Wearable Device

Bipolar disorder is a common chronic recurrent psychosis and it mainly relies on doctors’ experience to determine the patient’s condition currently. We aimed to find a useful methodology to diagnose the mental state and guide medical treatment by using speech signal processing. Methods: Firstly, the feature classes were extracted (e.g., Pitch, Formant, MFCC, GT). Secondly, class separability criterion based on distance (the Between-class Variance and Within-class Variance) was adopted as an evaluation criteria to get the features assessment, and then, we found LPC played a core role on the all features. According to the experiment, the SVM have a good performance for the single patient up to 90%, and the GMM classifier yields the best performance with a classification rate of 72% for multi patients. The newly proposed methodology provide a good method for helping diagnose bipolar disorder.