Artificial intelligence helped create a dog language translator device

Scientists have found a way to decipher the sounds made by dogs with the help of artificial intelligence trained to understand human speech. The research, published on arXiv and presented at an international conference, uses AI models such as Wav2Vec2 to decode dog barks. The basic idea is that these models can learn to recognize different aspects of a dog’s barking, such as the dog’s gender, emotion, and even breed.

To test the accuracy of the models, the researchers used two different data sets. One of them was trained in human language and then tuned to dog barking. According to the results of the study, a model that was first trained on human speech and then tuned to dog barking performed best in identifying the dog’s breed and emotions. It achieved 62% accuracy in breed and emotion recognition, 69% accuracy in determining a dog’s gender, and 50% accuracy in identifying a specific dog from a set.

The model, pre-trained on human language, was better able to understand and identify dog ​​barking, especially after being tuned to sounds made by 74 dogs, including 21 French poodles, 11 schnauzers and 42 chihuahuas. Researchers believe that using patterns derived from human speech is the key to deciphering dog language. They plan to further improve the models by using more breeds, emotions and species of dogs to expand the study.