Speech Emotion Recognition using Deep Learning

Abstract

Speech Emotion Recognition (SER) is an emerging field that involves recognizing emotions conveyed in speech. Emotions expressed through speech can greatly impact decision-making. This paper delves into the topic of speech emotion recognition (SER) and its focus on interpreting emotions conveyed through spoken language. The importance of SER lies in its potential to improve human-computer interaction, cognitive analysis, and psychiatric assessment. The study combines and preprocesses audio data from various datasets, such as RAVDESS, CREMA-D, TESS, and SAVEE, and uses log mel spectrograms to effectively extract features. Various methods including CNN models, and standard and optimized feature extraction techniques are used. The results suggest that SER has significant real-world applications and the approaches provided effectively identify emotional and voice signals.