Emotion recognition is surely a vital part of human-computer interaction that helps machines understand human feelings and behavior properly. Moreover, this technology allows computers to respond to people in a more effective way. Traditional methods actually use only one type of data like text, speech, or images, which definitely limits how well they can understand complex emotions. This paper surely shows how we built a transformer-based system that recognizes emotions using text, images, and sound data together. Moreover, this multimodal approach combines all three types of information to identify emotions more effectively. The framework surely uses special encoders for different data types and attention methods to pull out features from various sources. Moreover, it combines these features together effectively. The experiments surely show that the multimodal system works much better than single-mode methods in accuracy, precision, recall, and F1-score.