A Real World Dataset of Canine Emotions Using Video and Audio

Published: 24 September 2026| Version 1 | DOI: 10.17632/5g5hg9hrkv.1
Contributors:
Sankirthan Rao, Keerthan Sanil

Description

This dataset was collected as part of a research effort to understand and recognize the emotional states of dogs through their natural behavior and vocalizations. The purpose of this collection was to investigate the feasibility of using both audio and visual data to recognize various dog emotions and lay the groundwork for automatic algorithms to recognize dog emotions. This data set has been gathered from various localities of Mangalore Approximately 127 audio-visual dog recordings collected for the study of Canine Emotion Recognition. The dataset consists of 4 emotion classes of unlabelled video recordings of dog: Happy, Sad, Angry, and Relaxed., These emotions are identified and assigned labels through the application of various clustering algorithms.. The dataset will contribute to the research and development of machine learning and deep learning algorithms for understanding and categorizing the emotional state of dogs based on visual and auditory cues. The recordings are suitable for emotion recognition in multimodal, audio-based, video-based and emotion classification systems. This dataset can serve as a basis for analyzing the correlation between the dog's behavior, expression and emotions for emotion recognition systems that recognize and process emotions automatically.

Files

Steps to reproduce

1.Data Collection: Go to various locations where dogs are available and annotate what they do and how they communicate by taking notes and using a camera that can record video and audio. 2. Audio-Visual Recording: Record videos of canines in various emotional states such as Happy, Sad, Angry and Relaxed. 3. Data Organization: Group collected recordings by emotion, using the observed behaviour and available annotations. 4. Environmental Noise Reduction: Enhance the quality of the recorded audio by removing environmental and background noise, making the dog's vocalizations more audible. 5. Audio Extraction and Processing: Extract the available dog vocalisations from the recordings and use the processed audio data. 6. Dataset Preparation: Make the corresponding video and audio recordings to be used in the study of emotions in dogs and in multimodal machine learning experiments.

Institutions

Categories

Computer Science, Artificial Intelligence, Computer Vision, Data Science, Machine Learning, Data Analysis, Deep Learning

Licence