Dogs and cats have become important members of many families. However, humans cannot accurately understand the emotions or intentions behind animal vocalizations.
This research focuses on developing a deep neural network model that classifies dog barks according to emotional states. The study uses YAMNet, an environmental sound classification model, and trains it with publicly available datasets as well as bark recordings collected from a puppy. Dog vocalizations are categorized into emotional classes such as excitement and desire, loneliness and anxiety, communication, caution, and intimidation, with the goal of estimating a dog’s emotional state from its bark.
The model was fine-tuned using the collected audio data and evaluated for classification performance. Experimental results achieved an accuracy of approximately 80%, demonstrating high performance for several emotion categories. In addition, a web-based application was developed as an example of practical use, allowing users to upload dog bark audio files and view the predicted emotional classification results.
Future work aims to improve classification accuracy by incorporating not only bark sounds but also environmental information and individual characteristics of dogs.
References
TensorFlow: Transfer learning with YAMNet for environmental sound classification
https://www.tensorflow.org/tutorials/audio/transfer_learning_audio
