Deep Learning Approaches for Effective Human Computer Interaction: A Comprehensive Survey on Single and Multimodal Emotion Detection

Emotions play an important role in communication, given its complexity, emotion recognition research is difficult because people convey their emotions through a variety of modalities, such as language, voice, and facial expressions. These cues often complement each other. This study explores the recognition of emotions from visual, auditory and also textual information. The proposed system offers flexibility in predicting features and recognizing categorical emotions, even when only one type of data (either audio or video or text) is available. To accomplish this, the system leverages pre-trained models like Efficient-Net, LSTM and Wav2Vec. The study on the evaluation of model performance includes metrics such as accuracy and precision, offering a comprehensive view of predictive capabilities.