Playful barking, aggressive growling… what if artificial intelligence helped you better understand your dog ?

Playful barking, aggressive growling... what if artificial intelligence helped you better understand your dog ?

Understanding the nuances of barking could greatly improve how humans respond to dogs' emotional and physical needs. AleksandarNakic/Getty Images

To the ear, barks all sound the same. However, dogs make vocalizations to tell us a whole bunch of things. If owners often struggle to understand their little companion, an American-Mexican research team is looking towards artificial intelligence to make their lives easier.

Researchers from the University of Michigan are working in collaboration with the National Institute of Astrophysics, Optics and Electronics (INAOE) in Mexico, developing an AI tool that would interpret canine vocalizations. He would be able to distinguish playful barks from aggressive growls, and even identify characteristics such as the dog's age, breed and gender based solely on the sounds the dog makes. ;rsquo;animal.

For their work, the academics used pre-existing AI models of voice recognition and transcription. "Using speech processing models initially trained on human voices, our research opens new perspectives on how we can use what we've built so far in speech processing. speech to begin to understand the nuances of dog barking", says Rada Mihalcea, professor of computer science and engineering, and director of the AI ​​Lab from the University of Michigan, in a press release.

But when developing their artificial intelligence software, Rada Mihalcea and his colleagues were faced with a lack of acoustic data on canine vocalizations. They therefore decided to record the barks of 74 dogs of different breeds, ages and sexes, and have them analyzed by Wav2Vec2, an automatic speech understanding model developed by Meta.

Very encouraging results

The results were very encouraging: Wav2Vec2 not only succeeded in classifying these barks into four different categories but it also outperformed other models specifically trained on canine acoustic data, with a accuracy of up to 70%. "This is the first time that techniques optimized for human speech have been used to decode animal language. We have discovered that sounds and patterns derived from human speech can serve as a basis for analyzing and understanding the acoustic patterns of other sound sources, such as animal vocalizations, says Rada Mihalcea.

If this research work is still in its early stages, it offers a glimpse of a future where we will eventually be able to decode animal language. Good news for biologists but also for “dog parents”. Indeed, understanding the nuances of canine vocalizations could greatly improve how humans interpret and respond to dogs' emotional and physical needs. And therefore to increase the well-being of man’s best friend.

Add a Comment

Your email address will not be published. Required fields are marked *

(function(d,s){d.getElementById("licnt2061").src= "https://counter.yadro.ru/hit?t44.6;r"+escape(d.referrer)+ ((typeof(s)=="undefined")?"":";s"+s.width+"*"+s.height+"*"+ (s.colorDepth?s.colorDepth:s.pixelDepth))+";u"+escape(d.URL)+ ";h"+escape(d.title.substring(0,150))+";"+Math.random()}) (document,screen)