Use of prosodic speech characteristics for automated detection of alcohol intoxication

Abstract

In this paper we describe our methodology for automatic detection of speaker alcoholization. Our task is restricted to detection of considerable alcoholization (alcohol blood level 0.8 per mille), so that a two-class classification problem is to be solved. In particular, our attention is focused on the influence of the alcohol intoxication on the prosodical aspect of the spoken language. A new kind of signal intervals underlying the extraction of prosodic features (phrasal units) is proposed along with a method for their localization, which makes it possible to avoid the word segmentation of the speech signal as a preceding stage of the classification process. We also assess the utility of various prosodic features computed on such intervals for the task specified above. In our experiments on unseen data, we achieved classification rates of almost 69% when discriminating between alcoholized vs. not alcoholized speech.