Cognitone: Declassifying Vocal Emotion

Speech emotion recognition (SER) is a crucial field that aims to accurately infer human emotions from speech signals, facilitating improved human-computer interaction. Despite the challenges associated with annotating audio data and the subjective nature of emotions, SER is essential for enhancing communication between humans and machines. In this study, we propose a novel approach based on Convolutional Neural Networks (CNN) to enhance SER performance. By leveraging features like tones, pitches, and mel-frequency cepstral coefficients (MFCC), this CNN-based model achieves an impressive accuracy of 83.29%. This model surpasses existing systems by recognizing eight distinct emotions (angry, calm, disgust, fear, happy, neutral, sad, and surprised), thereby expanding the emotional spectrum captured by SER. This system is trained using a combination of the TESS and RAVDESS datasets and incorporates real-time audio recording functionality, making it a robust and versatile tool for real-world applications in speech emotion recognition.