Abstract

The speech-to-emotion detection deals with detecting emotions through voice. While having a face-to-face conversation with another person, it is often possible to gauge their emotion through cues such as expressions, body language, and more. However, during telephone conversations, it becomes challenging to grasp an individual's emotional state. This work is aimed at recognizing emotions through one's speech, indicating that emotions can be identified by analyzing tone and pitch. This also sheds light on how animals can recognize human emotions. In this study, we propose a framework to tackle this challenge. Firstly, we collected a diverse dataset of speech samples annotated with emotion labels. Next, we extracted relevant features from the speech signal, including pitch, energy, MFCCs, and prosodic features. Using this dataset and feature set, we trained machine learning models to classify emotions