In recent years, multimodal emotion recognition has gained significant attention due to its potential applications in various fields, including healthcare, education, and entertainment. This paper presents a novel multimodal emotion classifier that integrates video, audio, and text features to achieve robust emotion recognition. Notably, the proposed model is designed to be lightweight, making it suitable for real-time applications, such as emotion-aware human-computer interaction and video conferencing tools. Our model demonstrates performance comparable but slightly less than that of state-of-the-art (SOTA) approaches in the field, showcasing its ability to accurately identify emotions across diverse modalities. The architecture incorporates attention mechanisms to enhance feature representation while maintaining a streamlined design, ensuring that it can be deployed in resource-constrained environments without sacrificing much of classification accuracy. The findings suggest that this model not only meets the demands of real-time processing but also allows to keep it light-weight.