An Automatic Method for the Analysis of Pitch Profile in Bipolar Patients

Psychiatric patients affected by bipolar disorder experience a mood swing, often ranging from mania to depression. Investigating biomedical signals to detect physiological correlates of mood changes is a more and more debated issue. In this work we describe an automatic method to analyse prosodic features estimated from speech signals. In particular we explore pitch dynamics in voiced part of syllables using some features borrowed for Taylor’s tilt intonational model. However, the approach here proposed differs substantially from Taylor’s one in that the features are estimated from all voiced segments without performing any analysis of intonation. This method results in features that acquire a different meaning and can be estimated automatically without any labelling step. The suggested approach has been tested firstly on an emotional speech database. Then an analysis on speech samples acquired on psychiatric patients in different mood states is introduces and the results are discussed.