(Web Desk) - In a breakthrough study, researchers at the University of Michigan have successfully leveraged artificial intelligence (AI) to decode dog barks.
“Our research opens a new window into how we can leverage what we built so far in speech processing to start understanding the nuances of dog barks,” said Rada Mihalcea, head of the University of Michigan AI Laboratory.
The study also explored the possibility of using AI to identify a dog’s age, gender, and breed based on its bark.
The researchers believe that a dog’s bark is closely linked to its context, aligning with evidence suggesting that sounds made by other animals, like monkeys and prairie dogs, can be predicted based on situational context.
AI for canine communication
In this study, the researchers focused on identifying specific emotions in dog barks, such as aggression, normalcy, negative squeals, and negative grunts.
The AI model, known as Wav2Vec2, was trained on two different datasets: one consisting entirely of dog barks and another pre-trained on nearly 1,000 hours of human speech and then fine-tuned on dog barks.
Surprisingly, the model pre-trained on human speech outperformed the one trained exclusively on dog barks. This indicates that patterns and structures inherent in human language can serve as a valuable foundation for interpreting animal vocalizations.
The AI model was able to interpret the emotional state of dogs, differentiating between playful and angry barks, with an average accuracy of 70%.
Besides, the model identified a dog’s breed with 62% accuracy and gender with 69% accuracy.
“Advances in AI can be used to revolutionize our understanding of animal communication, and our findings suggest that we may not have to start from scratch,” added Mihalcea.
The team, led by U-M doctoral student Artem Abzaliev, repurposed existing computer models trained on human speech, due to the lack of a comparable database for dog vocalizations. They collected barks, growls, and whimpers from 74 dogs of various breeds, ages, and sexes in different contexts and fed them into a machine-learning model.
“This is the first time that techniques optimized for human speech have been built upon to help with the decoding of animal communication,” noted Mihalcea.
The results suggest that sounds and patterns from human speech can be used as a basis for analyzing animal vocalizations.