According to many researchers, science will soon be able to decode dog barking

In the news StadyMany researchers address the topic of dog barking and explore ways to understand it.

Using artificial intelligence, scientists at the University of Michigan are studying different types of bark in hopes of decoding them.

Rich communication

In the introduction to the study, the researchers explain their approach: “Like humans, animals make extensive use of verbal and non-verbal forms of communication, including a wide range of vocal signals. In this post, we address dog vocalizations and explore the use of self-supervised speech representation models, pre-trained on human speech, to solve canine bark classification tasks that find similarities to human-focused speech recognition tasks. »

See also:

As the study specifies, the vocal cues used by dogs will be highly contextual: so scientists hope they can classify the different types of barking using artificial intelligence so they can decipher the emotions of our canine friends.

To begin this analytical work, the researchers first trained their AI model by presenting them with two data sets of 1,000 hours each: one containing human discussions and barking, and the second only dog ​​barking.

Surprisingly, the AI ​​model trained using human babble combined with barking performed better than the model trained using only dogs' vocal cues. This may indicate that the structure of human discussions is a reliable basis for studying animal vocalizations.

During their analytical work, researchers from the University of Michigan were able to confirm that their artificial intelligence tool achieved an accuracy rate of up to 70% in distinguishing between aggressive barking and play barking.

Even if there is still a long way to go, researchers are convinced that these first discoveries are an interesting foundation that will one day allow us to understand with more precision what our canine friends have to tell us.