Artificial intelligence learns to understand barking dogs

Researchers from the University of Michigan are developing artificial intelligence tools that will help understand what expresses the barking of a dog, writes Techinsider.

Earlier, a scientific prize was established, which will be awarded to scientists who have made progress in creating tools for interspecies two-way communication.

Jeremy Koller, chairman of the foundation that established the prize, says, “Just as the Rosetta Stone revealed the secrets of hieroglyphics, I am convinced that the power of artificial intelligence will help us establish interspecies dialog.”

The work of scientists from the University of Michigan was done before the prize was established, but it fits into its format and is just based on the use of AI.

The main problem faced by scientists when training an AI model to decode dog vocalizations is a severe lack of raw data.

In the recordings of human speech, scientists do not experience any shortage, but the recordings of dogs to train the AI are not enough.

To solve the problem, the scientists used an AI model that was originally trained on human speech and understands several languages “by ear” several languages well.

This model was trained on a small database of dog vocalizations.

As the scientists say, the first results are encouraging. The model is confident enough to recognize and interpret the nuances of dog barking that people do not notice.

The AI successfully recognized specific dogs, identified the breed quite reliably, and determined the sex of the animal.

In addition, the model successfully linked vocalization to context; for example, it distinguished aggressive barking from playful barking.

The scientists say: This is the first time that methods optimized for human speech have been applied to decipher animal vocalizations.