The Breathalyzer in Every Hand
Smartphone Measures Blood Alcohol Level via Microphone
Smartphones can predict blood alcohol levels by analyzing the voice - with an accuracy of 98 percent. In the future, gait patterns and writing behavior could also be incorporated into the blood alcohol level measurement. The cell phone test is intended to prevent accidents based on drunkness.
»Please blow into the loudspeaker« - what sounds like a questionable idea in a drunken state could ensure greater road safety in the future. Researchers at Stanford University have developed a model that can be used to determine the blood alcohol level of drunk people using a voice test - AI-supported software on a cell phone predicts how much alcohol is in the blood.
Analysis of Voice Frequency and Pitch
At the beginning of the study by Stanford Medicine and the University of Toronto, the test subjects first had to read out short texts and tongue twisters while sober. They were then given a weight-dependent dose of alcohol in the form of vodka.
Subsequently, the study participants with a blood alcohol level of around 0.8 read further tongue twisters aloud every hour and up to seven hours after drinking, while a smartphone placed next to them recorded their voices. The researchers also measured breath alcohol every half hour.
To measure alcohol via the voice, the scientists isolated the individual voices, broke them down into one-second increments and analyzed measures such as frequency and pitch. When compared with the breath alcohol results, the changes in the participants' voice patterns over the course of the experiment were able to predict the level of alcoholization with an accuracy of 98%.
Signal processing meets machine learning
»The accuracy of our model really surprised me,« says lead researcher Dr. Brian Suffoletto, Associate Professor of Emergency Medicine at Stanford. »Our superior accuracy in analyzing speech characteristics during alcohol intoxication is due to the fact that we use state-of-the-art signal processing, acoustic analysis and machine learning techniques.«
In the future, anyone should be able to use their cell phone to spontaneously measure their own alcohol level. According to the researchers, this should prevent injuries and deaths resulting from car or other accidents. The almost ubiquitous presence of smartphones makes the devices an optimally available breathalyzer - the cell phone can show people how drunk they really are.
Integration of Further Sensors for Biomarker
Dr. Suffoletto assumes that alcohol monitoring will also combine several cell phone sensors in the future - for example, gait, voice and texting behavior. »Integrating tests with different sensitivities and specificities can increase overall performance,« he says. »Also, we can't always rely on users to continuously input data. A person may not speak for hours, but they are walking. Or they may be standing at a bar, neither walking nor talking, but actively texting.«
The language test currently only works for English-speaking people; the study now needs to be continued with different ethnicities. The researchers also want to recruit companies for their project that already have voice samples from smart speakers. Dr. Suffoletto also wants to convince the British NIH (National Institutes of Health) to set up a database for this type of digital biomarker. His overarching goal is to develop an intervention system that people can use intuitively and voluntarily to prevent injuries and save lives. (uh)
