Evaluating Multi-Layer Perceptron and Recurrent Neural Networks for Speech Emotion Recognition

Speech Emotion Recognition (SER) aims to automatically recognize and classify emotional states based on speech signals. This article provides an overview of SER, discussing challenges, techniques and applications. Relevant features including prosodic, spectral and linguistic features are extracted to accurately capture the emotional content. Classification algorithms such as traditional machine learning and deep learning approaches such as recurrent and convolutional neural networks are explored. Issues such as variability, lack of annotated data and cultural factors are addressed through strategies such as knowledge improvement and learning. SER finds applications in affective human- computer interaction, psychological research and emotion-awareness systems. The potential impact of SER on improving human-machine interaction and user experiences is highlighted. In this paper, we compared an MLP classifier with recurrent neural networks (RNN) for speech emotion detection.