Speech emotion recognition using diffusion map- based machine learning with classification technique

Speech is one of the most common and straightforward methods of communication, expressing emotions and identifying others’ sentiments. Emotional speech processing is a field of study that focuses on understanding and analyzing the emotional content of spoken language. Although the categorization of different emotions requires not only a need for professional data samples but also labeled data and large datasets with high computational time with less convergence. However, the problem of redundancy and high dimensionality caused the infeasibility of manual extraction of various thresholds of emotions. Such deficiencies include the problem being addressed by the research and its scope. This study proposes a speech emotion detection system that outperforms a current system based on data selection, extraction, and reduction of features to identify speech intuitions based on emotions more accurately by using the Diffusion Map (dimension reduction) method with an ensemble classifier (DT, KNN). The proposed Diffusion Map-based non-linear approach is then trained and evaluated on the famous speech database Emo-DB Berlin Dataset to extract speaker-independent features. The improvement of 3.1% result provides cutting-edge performance and a reliable framework for Speech Emotion Recognition (SER). The proposed method would also provide the gateway to understanding and assessing health issues and different diseases via images in clinical fields.