Researchers are using an AI model trained on human speech to decode the secret language of dogs. This research comes from researchers at the University of Michigan, the National Institute of Astrophysics, Optics and Electronics in Mexico, and the Optical and Electronic Research Institute. Promising research results were presented at an international conference last week, suggesting that today's AI models may hold the key to understanding animal language, at least to some extent. "There is still much we don't understand about the animals we share this world with," said Rada Mihalcea, director of the Artificial Intelligence Lab at the University of Michigan, in a press release. "Advancements in artificial intelligence can be used to revolutionize our understanding of animal communication, and our findings suggest that we may not have to start from scratch." The study utilizes state-of-the-art AI speech model Wav2Vec2 to identify the emotions, gender, and breeds of dogs behind any given barking sound. Researchers trained and compared the model using two different datasets: one trained solely on dog barks and another pre-trained on human speech and fine-tuned on dog barks. The model pre-trained on nearly 1000 hours of human speech recordings performed better. The researchers then fine-tuned the model on a dataset consisting of barks from 74 dogs, including 42 Chihuahuas, 21 French Poodles, and 11 Schnauzers. This AI model, trained on both humans and dogs, is able to identify the emotions of dogs with 62% accuracy, the breeds with 62% accuracy, the gender with 69% accuracy, and specific dogs from a group with 50% accuracy. All of these scores surpass those of AI models trained solely on dogs, indicating that sounds and patterns from human speech may serve as a foundation for understanding animals. When attempting to interpret the emotions behind dog barks, researchers hypothesized that the barks are related to their context. Existing evidence suggests that sounds emitted by monkeys and prairie dogs can be predicted based on their circumstances. In this study, researchers attempted to categorize some of the dog's emotions into aggressive barks, normal barks, negative screams, and negative growls. While dogs may experience a wider range of emotions, these noises were mostly present in their dataset. "By using a speech processing model initially trained on human speech, our research opens up a new window that allows us to leverage the knowledge established in speech processing so far to begin understanding the subtle differences in dog barks," said Mihalcea.