Automatic classification of dog barking using deep learning.

Area of Science:

• Animal Behavior
• Bioacoustics
• Machine Learning

Background:

• Dog vocalizations convey information about emotions and internal states.
• Intelligent audio analysis uses signal processing and machine learning for acoustic signal interpretation.
• Classifying bark characteristics can aid professionals interacting with dogs.

Purpose of the Study:

• To develop and evaluate a method for classifying dog barks based on identity, breed, age, sex, and context.
• To utilize deep neural networks for analyzing acoustic properties of canine vocalizations.
• To provide a foundation for technological advancements in understanding dog communication.

Main Methods:

• A three-stage methodology: pre-processing, characterization, and classification.
• Utilized deep neural networks (DNNs) for bark classification tasks.
• Trained and evaluated models using 19,643 barks from 113 dogs of diverse breeds, ages, and sexes.

Main Results:

• The proposed method demonstrated outstanding performance in classifying bark attributes.
• Achieved superior results compared to previous research in dog vocalization analysis.
• Identified relevant audio features and optimal DNN architectures for each classification task.

Conclusions:

• The developed method shows significant potential for analyzing and understanding dog barks.
• The findings provide a strong basis for future technological developments in canine bioacoustics.
• While not yet ready for ethological practice, the performance indicates a promising direction.