Leveraging Deep Learning to Predict Cat Emotions Using Audio

Cats are sentient beings which communicate through body language, scent cues, and most importantly vocalizations. Their emitted sounds represent a range of thoughts and emotions, which can be challenging to interpret. Recently, researchers have shown interest in studying the emotions of cats, especially in the context of analyzing their vocalizations. This paper describes a new approach that allows predicting cats’ emotions using audio data involving a layered architecture of two deep learning models “Kisha” and “Beti”. We leverage convolutional neural networks and highly performant computer vision models such as Resnet by employing spectrogram pre-processing. Both models are trained on a curated dataset of over 3,000 audio clips sourced from datasets and online platforms. Beti, which is capable of distinguishing between different types of cat sounds achieves an accuracy of 85%, while Kisha, capable of predicting the emotion of a cat’s meow reaches 91.8% accuracy. Furthermore, we present “CatMotion”, a mobile application, which enables real-time emotion analysis of cat sounds via a user-friendly interface. Our findings demonstrate the potential of artificial intelligence to enhance human understanding of feline emotions, with wide applications, including pet care and animal behavior studies.