Speech Emotion Recognition Using Machine Learning

Abstract

Speech has a significant role in sharing thoughts as an emotion-carrier. Speech emotion recognition has numerous applications in areas such as robots, mobile services, and psychological testing. Although the recognition of human emotions from speech has advanced thanks to deep learning, there are still issues with SER research, including a lack of data and poor model accuracy. When using Speech Emotion Recognition (SER), emotional traits frequently show up as various energy patterns in spectrograms. The potential applications of Speech Emotion Recognition (SER) in fields including human-computer interaction, emotional computing, and psychology have led to the field's emergence as a significant study area in recent years. The goal of this study is to investigate how to extract features from voice signals and identify emotions using MFCCs, and standard neural networks (CNNs). In the proposed approach, voice data is preprocessed to reduce background noise and extract pertinent features like MFCCs. The data are obtained from the Kaggle open source RAVDESS and are then reduced in dimension using feature selection technique. The dataset size is then increased and the robustness of the model is improved using data augmentation approaches. Due to its versatility and success in a variety of classification problems, the CNN algorithm is chosen as the classification approach.The outcomes demonstrate that the suggested approach outperforms other cutting-edge techniques and achieves excellent accuracy. The proposed system illustrates the value of data preprocessing, feature extraction, and classification techniques in achieving high performance in this task and shows the possibility of using MFCCs and CNNs for speech emotion recognition.