Glancing through the latest edition of the Journal of the Acoustical Society of America, I was struck by this paper from Heinrich and Schiel, ‘The influence of alcoholic intoxication on the short-time energy function of speech‘. We’ve all witnessed peoples’ speech becoming slurred when they’re drunk. But can a computer pick up those tell-tell signs, and detect whether someone is intoxicated from their speech?
A number of scientific studies have explored how drinking alcohol changes speech. The average fundamental frequency of the voice (or pitch) tends to go higher, but the pitch also varies more. This rise in frequency might be caused by people talking louder when they are drunk. Another effect is the slurring of speech or what is called disfluency. As people get drunk, they are more likely to stumble over their words, repeat them, elongate them, or even miss out some of the words.  However, there are who do not show a significant change in their speech when intoxicated. Some people appear to be able to hide how inebriated they are.
Can a computer detect drunkenness? The most successful system I could find in the literature could detect intoxication correctly about 77% of the time. It did this by looking at a number of features of the speech that could be extracted from a recording by a computer. Some of these related to frequency and others to the rhythm . While a 77% success rate might seem high, this means 23% of cases are misidentified by the computer. Coincidentally, this is about as good as a human can do. When scientists have tested peoples’ abilities to spot drunks by listening to speech alone, the success rate isn’t any better, and sometimes worse .
Maybe the solution is to use a multi-modal approach as is being tried in speech recognition. To combine the speech features with other non-audio signs of drunkenness. Maybe one of these features could be physical clumsiness, as ably demonstrated by Les Patterson in the video. One of the reasons for research into detecting intoxication, is to put systems into cars that detect drunk drivers and stop them using their vehicle. The car would be voice activated and would only work if the driver was judged to be sober. Maybe alongside listening to the speech, the car needs to have accelerometers in the driver’s seat, spotting those who have enough alcohol in them to dull their motor skills used when sitting.
Wwwwhat do youuuu think? Please, pleeeeeaaasee comment below.
 H. Hollien, G. DeJong, C. Martin, R. Schwartz, K.Liljegren, “Effects of ethanol intoxication on speech suprasegmentals”, J. Acoustic. Soc. Am. 110 (2001) 3198-3206.
 Florian Schiel, Christian Heinrich, Veronika Neumeyer, “Rhythm and Formant Features for Automatic Alcohol Detection”. Proc. INTERSPEECH 2010, ISCA, Makuhari, Japan, pp. 458-461, 2010.
 Björn Schuller et al. “Medium-term speaker states—A review on intoxication, sleepinessand the first challenge” Computer Speech and Language 28 (2014) 346–374