Automatic Discrimination of Domestic Cat Sounds and Imitations

This study presents the results of automatic discrimination to distinguish the sounds of domestic cats and the sounds of domestic cats imitated by humans. In this work, our main contribution is domestic cat sound imitation data set creation and investigation of classification accuracy both between original cat sounds and imitations mutually. First, we analyzed the specific types of vocalization of cats, which are the main representatives of the cat sound categories. Then, we use Freesound to create a data set of these specific type of cat sounds vocalizations. We created an imitation data set consisting of the vocalizations of the humans precisely for the same categories. We use Librosa to extract the features of these new data sets, and Weka to distinguish cat sounds from their imitations by using machine learning techniques.