COMPARATIVE ANALYSIS OF ML AND DL TECHNIQUES FOR SPEECH EMOTION RECOGNITION SYSTEMS

Virtual assistants, powered by Artificial Intelligence (AI), increasingly incorporate Speech Emotion Recognition (SER) to enhance user interactions.By recognizing and responding to emotions, these assistants provide personalized and effective support across various applications.SER, a fast-growing area, detects emotions through features like pitch, intensity, and Mel-Frequency Cepstral Coefficients (MFCCs).SER systems use machine learning and Deep Learning techniques, such as multilayer perceptron classifiers, to predict emotions like happiness, sadness, fear, surprise, neutral, disgust, clam, and anger.This is essential for enhancing humancomputer interaction and assessing mental health.This Paper explores the design and improvement of a Speech Emotion Recognition System with various kinds of deep learning and machine learning methods and multiple datasets.The main goal is to compare the effectiveness of multiple ML and DL techniques to improve the accuracy.This Paper implements a comprehensive experimental setup encompassing data preprocessing, feature extraction, model training, and evaluation, focusing on model accuracy.The analysis, illustrated through various graphs and charts, indicates that both traditional ML and DL models can be improved.Advanced techniques such as ensemble methods, and data augmentation were applied, resulting in significant accuracy enhancements across all datasets.The findings provide valuable insights into the comparative performance of different algorithms and establish a robust framework for future SER research.