Emotion Recognizer for Speech Using Deep Learning Techniques

In the realm of human experience, the recognition of emotions has gained increasing significance, shaping the way we interact with technology and understand ourselves. This study delves into the domain of emotion recognition using audio, presenting the EmotionNet model. Leveraging the Toronto emotional speech set (TESS) dataset, which spans seven diverse emotional classes, the EmotionNet model aims to decode the intricate tapestry of human emotions. Through the application of a Long Short-Term Memory (LSTM) based model, this study seeks to capture the temporal nuances embedded within speech patterns that correspond to various emotional states. This study showcases the remarkable potential of the EmotionNet model, achieving an impressive 99.46% accuracy and an outstanding Area Under the Curve (AUC) of 99.90%, enhancing its status as a very powerful tool for speech analysis-based emotion recognition. This study makes important strides in the field of emotion identification and has broad ramifications for both psychology and technology.