Human perception of alcoholic intoxication in speech

In this paper we describe a perception experiment on intoxicated and sober speech of 161 speakers recorded in the German Alcohol Language Corpus. 72 listeners achieved an average discrimination rate of 63.1% when asked to choose from pairs of stimuli which one sounded intoxicated. Perception results were not gender-dependent and no hidden effects were found in a control group test. Since earlier studies reported higher fundamental frequency for intoxicated speakers, the influence of fundamental frequency as a potential acoustic cue in human perception of intoxication was analyzed. Results show a significantly higher detection rate for speakers who produce a higher fundamental frequency when being intoxicated, and a higher success rate for listeners who show a general preference for choosing the stimulus with higher fundamental frequency. However, human listeners do not consistently exploit this acoustic cue, since a simple algorithm which always classifies the stimulus with higher fundamental frequency as intoxicated would lead to a better performance of 82% discrimination rate.