The secret language of dogs is an intriguing thing. I’ve long wished I could understand the little grunts and barks my corgi sends my way each day. Sure, we can come up with our own interpretations, but what do those sounds really mean to the dogs making them? Well, AI could soon help us figure that out.
The researchers involved in this new study, which is currently available on arXiv and was presented during an international conference, say that AI models like Wav2Vec2 can identify a dog’s gender, emotion, and breed based on the bark. To test the accuracy of the models, they used two different sets of data to train the AI models and then compared the results.
One was trained on human speech and then fine-tuned with the barks, while the other was trained only on the barks. They found that the model pre-trained with human speech did better at understanding and identifying the secret language of dogs. From there, they further fine-tuned it with barks from 74 dogs, including 21 French Poodles, 11 Schnauzers, and 42 Chihuahuas.
Overall, the model trained on humans and dogs was able to identify the breed and emotion with 62 percent accuracy. It was able to identify the gender 69 percent of the time and could even identify a particular dog out of a set 50 percent of the time. This model did much better than the AI model, which was trained only on dogs.
The researchers say that this suggests that patterns derived from human speech could serve as a solid foundation for helping us decode the secret language of dogs. The researchers want to test the model using more breeds, emotions, and species just to discover how far the finding extends. From here, they hope to build up and find a way to understand what those barks mean.
But, if these findings are correct, we could be moving forward in our attempts to finally understand what dogs are thinking and saying. And, who knows, maybe one day, all our dogs will have those collars from Up, and we’ll be able to communicate with our furry best friends with ease.
