Barking with meaning: to understand the language of dogs, researchers train AI on humans.

This is how artificial intelligence recognizes the emotions of animals much better.

AI models trained in human language can be transformed to better understand dogs. This conclusion was reached by a group of scientists led by specialists from the University of Michigan. The promising results of their research show that AI could be the key to understanding animal languages, which would revolutionize science.

The study used the Wav2Vec2 artificial intelligence speech model, which allows you to identify the emotions, gender and breed of a dog by barking. To train the model, the scientists used two data sets and compared their performance.

In one case, the AI ​​was trained from scratch only to hear dog barks, and in the other, it was pre-trained to perceive human speech, then reconfigured to use dog language. And the model, trained on almost a thousand hours of human recordings, showed significantly better results. Both models were also refined using the barking sounds of 74 dogs, including Chihuahuas, French poodles and schnauzers.

The AI, which used human speech and dog barking, identified dog emotions with an accuracy of 62%, breed - 62%, gender - 69%, and identified a specific dog from the general group with a result of 50%. All scores are higher than those of a model that was trained only in dog language.

Researchers believe that patterns and individual sounds and intonations identified by AI in human speech can serve as the basis for understanding animal language. This expands the use of artificial intelligence systems that may not have originally been designed to recognize the nuances of dog communication.

In the future, the researchers plan to test more dog breeds and also try to apply the model to other animal species. This will allow us to evaluate the potential of the technology.