Scientists create a device that translates dog language using artificial intelligence

Scientists have found a way to decipher the sounds of the secret language of dogs using artificial intelligence that has already been trained to understand human speech.

Here's What We Know

This study, available on arXiv and presented at an international conference, uses artificial intelligence models such as Wav2Vec2 to decode the language of dog barking. The basic idea is that these models can learn to identify different aspects of dog barking, such as gender, emotions, and even the breed of dog.

To test the accuracy of the models, the researchers used two different datasets. One was trained with human speech and then tuned using barking.

According to the results of the study, the model that was first trained with human speech and then tuned with dog barks performed better in identifying the dog's breed and emotions. It achieved 62% accuracy in identifying breed and emotion, 69% accuracy in identifying dog gender, and 50% accuracy in identifying a specific dog from a set.

The model pre-trained with human speech was better at understanding and identifying dog language, especially after being trained with the barking of 74 dogs, including 21 French poodles, 11 schnauzers, and 42 Chihuahuas.

This proved to be better than a model trained on dogs alone.

The researchers believe that using patterns derived from human speech is the key to unlocking dog language. They plan to further improve the models using more breeds, emotions, and types of dogs to expand the study.

Source: Michigan News