Just one year ago, a group of researchers from the University of Michigan surprised the world with an experiment that showed that dog barking is not as chaotic as it seems. That first foray, published in mid-2024, was able to get an AI model to identify common vocal patterns across different genders and situations. Today, the same team is going one step further with a new study that refines and digs deeper into dogs’ vocal behavior.
The goal remains the same: to harness the power of sound processing models developed for humans and adapt them to the analysis of dog sounds. However, the project is now working with a much larger volume of data and with Deep learning More developed and achieved accuracy This exceeds the results obtained in the initial study.
Expanding the database and methodology
In this second stage, the researchers collected 19,643 barks from 113 dogs Of different races, ages and genders, recorded in natural environments, whether domestic or professional. Each bark was categorized according to the context in which it occurred: play, alertness, excitement, stress, or loneliness. This broad repertoire allows AI models to not only recognize public sentiment, but also recognize it Specific patterns associated with individual characteristics From every dog.
The team used deep neural networks to train classification models at three levels. First, processing the data to adapt it to the format required by the models, second, describing the barks through vocal representations that improve pattern identification, and third, classifying the information through hyperparameter adjustments that increase the accuracy of the system.
The dog’s identity, breed, and sex
One of the most notable developments in the study is that artificial intelligence can distinguish between individual dogs, in addition to classifying their breeds and genders. This ability has opened the door to practical applications in shelters, dog education and animal monitoring ever since Every dog has a unique vocal profile This reflects both their physiology and mood.
The accuracy achieved in individual identification, although still not perfect, greatly improves the results of the first study, which focused primarily on the emotional interpretation of barking.
Context and emotion
Analysis of barking is not limited to identifying the dog that is barking, but the researchers were able to link each sound to the context of the emission and the sound. Emotional state clear. For example, barking during play is clearly distinguished from alert or stress barking. This is possible thanks to the fact that models Deep learning It captures the nuances in frequency, duration and structure of the sound emerging from the human ear, providing a more accurate interpretation of an animal’s internal state.
The future of the project
In the long term, researchers aim to create Portable systems that interpret barking in real timeDetect signs of pain, anxiety or arousal, through applications in homes, veterinary clinics and working environments with dogs. As in the previous phase, the team stresses that the goal is not to “translate” barking into human language, but rather to better respond to the emotional and physiological needs of dogs.