Listening to Feelings of Emotion: Exploring ML and DL for Speech-Based Detection and Classification

Human emotions are pivotal in communication, and their recognition holds significance across various domains like human-computer interaction and mental health monitoring. The emotions are usually expressed in terms of facial and the speech. This paper presents a speech-based approach to emotion recognition, particularly focusing on discerning genuine from insincere emotions. By leveraging information from multiple modalities, the system aims to enhance accuracy and robustness in emotion recognition, employing a blend of machine learning (ML) and deep learning (DL) techniques. Utilizing algorithms such as DT (Decision Tree), KNN, SVM, GMM and LSTM for speech expression recognition enables the system to extract quantifiable features, thereby improving its capability to distinguish between genuine and fake emotions. Evaluation involves testing with diverse datasets and performance metrics, particularly emphasizing the system's efficacy in detecting fake emotions. The proposed model is applied mainly on the Indian English Speech. The proposed system is able to identify overall 84%, 98.9%, 58.5%, 91%, 99.29% of the accuracy over the DT, SVM, GMM, KNN and LSTM models.