Machine Learning and Signal Processing for Speech-Based Sentiment Detection

Abstract

Speech-based mood analysis is a way to figure out how someone is feeling just by listening to what they say. It uses machine learning and signal processing. Fixing issues with background noise, vocal intonation, and pitch change will make mood classification much more accurate. The suggested method creates measures like MFCC, pitch, energy, and spectral characteristics after changing the audio samples' volumes and cutting down on noise. Next, random forests, deep learning methods, and support vector machines (SVMs) are used to put the features into groups. The experiment results show that combining signal processing with machine learning makes sentiment prediction much more accurate. Deep learning does better than other methods. This technology could be used for many things, like making user interfaces, keeping track of medical conditions, and analyzing customer comments.