Not Just Woofs Anymore: How AI Might Decode Dog Vocalizations

Have you ever been stumped about what your dog wants to say? Researchers at the University of Michigan, in collaboration with Mexico’s National Institute of Astrophysics, Optics and Electronics (INAOE) Institute in Puebla, are weighing the use of AI to develop tools that can identify whether a dog’s bark is a playful one or a show of aggression.

New Guinea
New Guinea singing dog, Credit: Wikimedia/R.G. Daniel

The AI tool can also shed light on other features like sex, age, and dog breed. Initially programmed for human speech, the model can become the first step taken toward training systems to decode animal communication. AI models can revolutionize communication with animals, and researchers do not need to start from scratch, says Rada Mihalcea, the director of the AI lab at U-M. Sampling enough animal calls to train the model would have taken time. To tackle these challenges, the team repurposed a model trained to analyze the nuances of human voices, like tone, accent, and pitch.

The team used a dataset of 74 dogs of different breeds, ages, and sexes and used the data to modify a human speech-analyzing model called Wav2Vec2. The model accurately classified the dog vocalizations 70% of the time.

Usually, dog handlers bank on various body language cues and other behaviors to understand what the dogs are thinking. Communication through signs and verbal cues forms a significant part of dog training. However, most dog owners fail to reach the control and discipline a dog handler could assert. The AI model can be seen as a bridge between humans and dogs.

Copyright @smorescience. All rights reserved. Do not copy, cite, publish, or distribute this content without permission.

Join 20,000+ parents and educators
To get the FREE science newsletter in your inbox!