Machine Learning Approach for Detection of Speech Emotions for RAVDESS Audio Dataset

The most effective technique to convey one's thoughts and actions to another is through the use of emotion. Emotion recognition from one's own voice is the most urgently needed technology right now. Emotions are attached to everything that a person owns. Every customer's feelings might provide light on what they really need from the customer service professional. Since human communication relies heavily on the ability to read one another's emotions, speech emotion recognition is crucial for understanding another person's inner workings. The time and frequency-based acoustic characteristics integrated with machine learning classifiers that can categorize emotions to enhance the performance of the proposed intelligent system. This research makes use of the audio records available in the Ryerson Audio-Visual Database of Emotional Speech and Song (RAVDESS). Emotion is gleaned from speech via temporal and spectral characteristics. Three machine learning classifiers, the SVM, DT, and RF classifiers, are then trained on the retrieved data. The RF classifier achieved 88.78% higher accuracy rate than the state-of-the-art approaches.