A method has been developed to identify drunk people by analyzing their voices

Tongue twisters can determine how much vodka a person has drunk

Researchers from Canada asked a group of study participants to read a tongue twister before drinking alcohol, and then every hour for seven hours after that. After analyzing these recordings, they developed a machine learning system capable of determining with 98 percent accuracy whether the person reading the text was under the influence of alcohol.

Photo: freepik.com

According to experts, alcohol is one of the oldest psychoactive substances. Historically, it was used to purify drinking water, such as to make grog by 18th century sailors, as a disinfectant, and even as fuel for cars and rockets, among other uses.

However, excessive alcohol consumption can lead to various negative health effects. In the short term, this can cause intoxication, characterized by mood swings, sluggish reactions, poor coordination, and slurred speech. Long-term and excessive alcohol consumption can lead to addiction, liver disease, cardiovascular diagnoses, and an increased risk of certain types of cancer, such as liver, mouth, and throat cancer.

In a new study, Brian Saffoletto and his colleagues noted that there are currently no commercially available tools to effectively detect alcohol intoxication. Devices such as transdermal blood alcohol monitors and portable alcohol meters can accurately determine blood alcohol content, but they are often expensive and inaccessible to the average user.

On the other hand, everyone knows that alcohol changes speech, which can be easily recorded by a microphone using mobile phones or voice recorders. Thus, the analysis of voice samples could be a very simple and effective way to detect alcohol intoxication.

The study included 20 adults, but the analysis was carried out on 18 participants because two of them did not provide voice samples. The average age of these participants was 29 years, of whom 72% were men.

On the day of the study, participants arrived at the laboratory at 08:00 am. Each person was asked to read aloud a tongue twister that was recorded using a mobile phone. After recording, respondents drank enough vodka mixed with lime juice and syrup to achieve a mouth alcohol concentration greater than 0.20 ppm. Participants had an hour to drink this alcoholic drink. Subsequently, every half hour for seven hours, the researchers measured the participants' blood alcohol levels and recorded them reading the tongue twister.

They used the recordings to develop a machine learning model to predict alcohol drinking. The final model was 98 percent accurate in predicting alcohol intoxication. This demonstrated that intoxication could be recognized through the analysis of voice recordings.

«In this laboratory study, we found that brief English speech samples were useful for classifying intoxication states in adults. To move the science of remote detection of alcohol intoxication forward, a much larger number of participants with more diverse vocal samples collected before and during the ascending and descending curves of alcohol intoxication is urgently needed,” — concluded Brian Saffoletto.

However, critics point out that the study used a small sample of English-speaking people who collaborated with the researchers. Results may be different if there are more respondents and a more diverse sample. It's also worth considering, skeptics say, that participants may have tried to hide their intoxication, as is often the case when measuring alcohol intoxication in real-world settings.