Model ‘trained’ by humans can accurately identify the meaning of dogs’ barks – Executive Digest

A new study has revealed that an artificial intelligence (AI) model trained on human data can more accurately identify the meaning of dogs’ barks. This is An interesting finding was presented In Joint International Conference on Computational Linguistics, Language Resources and Assessment.

Imagine being able to understand your dog’s every bark, whine or growl. It’s the focus of a recent study looking at how AI interprets dogs’ vocalizations, distinguishing between playful barks and aggressive growls, and identifying characteristics such as a dog’s age, breed and gender.

The team of researchers collected a dataset of barks from 74 dogs between 5 and 84 months of age in Tepic and Puebla, Mexico, and were mainly Chihuahuas, French Poodles and Schnauzers. Recordings were made in the dogs’ natural home environment to capture authentic vocal responses.

Dogs were exposed to a variety of stimuli designed to elicit different types of vocalizations, such as the presence of a stranger, playful interactions, and simulated attacks on the owner. Voices were captured with a Sony CX405 Handycam camera, and only the audio components were used for analysis.

Audio clips were divided into short segments and manually annotated based on the context in which they occurred. This procedure produced 14 different types of vocalizations, namely very aggressive barking at a stranger, normal barking at a stranger, and playful barking during games.

Central to the analysis is the use of a sophisticated AI model called Wav2Vec2, initially developed for human speech recognition. As PsyPost reports, the researchers fine-tuned the model by investigating tasks such as identifying individual dogs, determining breed, predicting sex, and associating barks with their specific environments, along with a dataset of dog vocalizations.

The AI ​​model demonstrated a remarkable ability to identify individual dogs based on their barks, achieving nearly 50% accuracy, compared to 24% for a model trained only on dog vocalizations. This fact, prior training with human speech provides a solid basis for understanding the complex structures of animal vocalizations.

Additionally, the model successfully identified a dog’s breed from its bark, with an accuracy rate of 62%. This result indicates that different dog breeds have unique vocal patterns similar to the accents in human speech. However, predicting a dog’s gender based on its vocalizations has proven more difficult, indicating that gender-related vocal cues may be less distinct.

Finally, the AI ​​model was good at distinguishing between different types of bark and landing barks in their specific environments. This fact highlights the advantages of using pre-human speech training to understand animal vocalizations.

Although the study results are promising, the researchers stress the need to include a wider range of dog breeds and more diverse samples in future research. This will ensure that AI models generalize more effectively to different canine populations.

This breakthrough represents a significant step forward in understanding dog behavior and communication, which could revolutionize the way we interact with our pets.