Speech Emotion Recognition on RAVDESS Dataset using Deep Learning

In the realm of Machine Learning, integrating state-of-the-art technologies is crucial for enhancing the efficacy of applications, particularly in Speech Emotion Recognition (SER), where cutting-edge technologies can substantially enhance performance. This paper investigates the utilization of Multilayer Perceptron (MLP) and Convolutional Neural Network (CNN) architectures in SER, aiming to surpass traditional method-ologles. MLP, renowned for its capability to capture intricate patterns, and CNN, adapted for analyzing spectrogram-based representations of audio data, are juxtaposed. The study involves a comparative analysis of MLP and CNN, emphasizing feature extraction, data processing, and model architecture. Results indicate that the CNN model surpasses MLP in capturing spatial hierarchies within audio features, thereby improving emotion recognition from speech signals. The CNN MLP achieves an accuracy of 86.4%, showcasing superior capabilities in discerning spatial hierarchies within audio features, while the CNN model, with an accuracy of 81.52%, lags behind in comparison to the CNN's performance.