Beyond Words: Extracting Emotions from Speech with AI Techniques

Emotions are important in daily life because they affect how individuals act, think, and react in a variety of circumstances. Thanks to current advances in Artificial Intelligence (AI), we can train machines to effectively understand and predict human emotions. AI based on; Speech Emotion Recognition (SER) helps in human-computer interaction. Devices like Smart Home Assistants can provide useful feedback thanks to the voice emotion recognition technology and the need to stay at home for a long time. This paper's main objective is to use machine learning and deep learning methods to analyse speech's emotional states. Random Forest, 1-D, and 2-D Convolutional Neural Network (CNN) are used to construct a high-quality voice emotion identification system. This method identifies seven different emotions: neutral, joyful, sad, furious, fearful, repulsed, and surprised. a standard dataset for emotion categorization used for training and testing that combines the Ryerson Audio-Visual Database of Emotional Speech and Song (RAVDESS) and Toronto Emotional Speech Set (TESS). To extract features from our sound wave data, we worked with Mel-frequency cepstral coefficient (MFCC) and Mel Spectrogram techniques using a python audio processing library called Librosa. 85.6%, 96.6% ± 1.5% and 98.4% ± 1.43% classification accuracies were obtained using Random Forest, 1-D CNN and 2-D CNN respectively. This research concludes that, 2-D CNN classification model can be effectively used to train machines to understand and predict human emotions.