Researchers have developed an AI model capable of deciphering dog barks with 70% accuracy, identifying emotions like joy, anxiety, hunger, and frustration. The model, Wav2Vec2, originally designed for human speech analysis, was adapted to study canine vocalizations.
Developed by Artem Abzaliev from the University of Michigan in collaboration with scientists from INAOE, the AI analyzes acoustic patterns in barks, howls, and growls in relation to a dog's body language. The study involved 150 dogs of various breeds in different situations.
Key findings include:
The AI can distinguish between barks indicating immediate needs and those representing more complex emotions.
Wav2Vec2 analyzes frequency, intensity, and rhythm variations in vocalizations.
The AI can infer a dog's age, breed, and sex based on its barks.
This technology opens new opportunities for improving human-dog interaction, studying animal behavior, and aiding in conservation efforts. Future challenges include expanding the database with more dog breeds and analyzing vocalizations in diverse situations to improve accuracy and extend the technology to other animal species, such as cats.