MULTIMODAL EMOTION RECOGNITION USING DEEP LEARNING AND FUSION TECHNIQUES
Loading...
Files
Date
2024-04-20
Authors
Mukhametsharip, Zhanna
Khamitova, Ainur
Kabdrakhmetova, Zhazira
Nurmakhan, Temirlan
Journal Title
Journal ISSN
Volume Title
Publisher
Nazarbayev University School of Engineering and Digital Sciences
Abstract
Emotion recognition plays a crucial role in human-computer interaction, significantly influencing the advancement of virtual assistants, mental health diagnosis tools, and customer experience analysis systems. Our senior project aims to develop an advanced multimodal emotion recognition (MER) model using modern deep learning techniques and fusion methods.
Most traditional emotion recognition models rely on a single modality for decision-making, such as facial expressions or text. However, this approach can be limited in capturing the complexity of human emotions. To overcome this limitation, we will integrate multiple input types to create a more comprehensive model, reducing misclassifications and improving overall system performance.
Our system includes an emotion recognition model and a user interface for interaction. The web application will serve as the interface, allowing users to upload video materials of a specified duration. The application extracts audio, video, and text from the uploaded video and feeds them into different deep-learning models customized for each modality. The outputs, representing probabilities for various emotion classes (e.g., ”happy,” ”sad,” ”fearful,” ”surprised,” ”angry,” ”disgusted,” and ”neutral”), will be combined using fusion techniques for enhanced accuracy. The web app then presents visual representations of the emotions through graphs and descriptions for user interpretation.
Description
Keywords
Type of access: Embargo, multimodal deep learning, emotion recognition
Citation
Khamitova A., Mukhametsharip Z., Kabdrakhmetova Z., Nurmakhan T. (2024). Multimodal Emotion Recognition Using Deep Learning and Fusion Techniques. Nazarbayev University School of Engineering and Digital Sciences