Multi-resolution classification techniques for PTSD detection from audio interviews

We describe a set of novel techniques used for identifying PTSD in patients from audio interviews. In particular, to improve classification performance, we utilize a multi-resolution decomposition of the audio signal into coarser and finer scales. We describe a simple ensemble probability averaging method together with a bagging approach, both of which outperform the classification scheme based on only the features of the untransformed signal. In addition, we describe our segmentation strategy and simple filtering techniques which can aid classification performance for audio recordings obtained in non-standardized recording conditions.