sciencetech-blog

comments

Up to 40 per cent of road deaths around the world are thought to be caused by alcohol.

But one day, cars could detect whether a driver's voice sounds drunk, to stop them from driving and prevent accidents.

Researchers in Germany have created the first library of drunk speech patterns, which could be used to predict drunkenness by listening for vocal cues. 

Researchers in Germany have created the first library of drunk speech patterns, which could be used to predict drunkenness in cars (Illustrated with a stock image)

Called the Alcohol Language Corpus, the database was made between 2007 and 2009 by giving people too much alcohol to drink, before recording conversations conducted in a stopped car, where the drunk person was sitting in the passenger seat.

There are currently conversations from 162 German men and women in the publicly-available audio database.

Experts at Ludwig Maximilian University of Munich and the Institute of Legal Medicine in the city collected the data, FastCompany reported.

From this, an algorithm that is capable of detecting whether someone is drunk by listening in on slurring speech patterns, has been created by computer scientists at Queens College and Columbia University.

Scientists have used the library to create an algorithm that can predict if a driver is drunk. The hope is that it could be built into cars to stop people putting their key in the ignition and driving off. The technology would act as a preventative measure, unlike breathalysers (pictured) that are used by the police to prove an offence

They found that they could detect drunken speech patterns much like identifying an accent.

Drunken speech hallmarks include stammering and stuttering. Voices also tend to rise in pitch and words begin to slur, they said.

It is hoped that the software could be used alongside devices such as locks and immobilisers to make it impossible for people to drive when drunk – perhaps by asking them a few questions while analysing the driver's voice.

'The cars themselves could listen to the driver, detect that the potential driver is intoxicated, and prevent the car from starting,' the researchers explained.

However, the safety system may be a few years away, because the algorithm to detect drunkenness is typically only right around three quarters of the time.

'That means, you want to go and drive your car, one in three times we're going to say you're drunk. That's bad for a product,' Professor Andrew Rosenberg, of Queens College New York (CUNY) said.

The next step is to add more languages to the library so that more experts can use the public library for their own research.