A Deep Learning Model for Speech Emotion Recognition using Multilayer Perceptron Classifier

Abstract: Speech signal processing research is still ongoing. It is the most effective method of Human-ComputerInteraction (HCI) and the primary means of human-to-human communication. Speech Emotion Recognition (SER)and other assessments of human behavior and emotion recognition from speech signals are expanding the field ofhuman-computer interaction (HCI) with a wide range of real-time applications. High-level deep features and otherobvious, discriminative information are incorporated into feature learning, which is essential to a SER system'sefficient operation. For the best prediction, we suggested the multilayer perceptron (MLP) classifier in this work.The MFCC, Chroma, and Mel characteristics are taken into account in this study. The main methods for featureextraction in this work were Mel Frequency Cepstral Coefficients (MFCC), Chroma, and Mel spectrograms.Librosa's built-in features are used to derive these features, which are then arranged into arrays for more in-depthclassification and analysis. We trained our suggested system and assessed its prediction ability using benchmarkRAVDESS emotional speech corpora, obtaining 87.9% recognition rates. The effectiveness and value of thesuggested system are demonstrated by its performance.