Speech Emotion Recognition Through Extraction of Various Emotional Features Using Machine and Deep Learning Classifiers

Identification of emotions along with the semantics from the speech signals is still one of the most challenging tasks because of the arising difficulty in the extraction and selection of the appropriate emotion features. The speech emotion recognition (SER) system has become one of the active research topics in recent years. Many researchers proposed various models to enhance the machine responses during the man–machine interactions by identifying the emotional state of the customer. This research work aims to propose a multi-class classification model to extract and classify the emotional state of the speaker into eight classes, namely happy, sad, surprised, disgust, neutral, calm, angry, and fearful. In this research work, two machine learning classifiers, namely k-nearest neighbors and random forest, are used along with the multi-layer perceptron classifier of deep learning for determining the emotional features from speech signals. The dataset for this research is obtained from the RAVDESS. Along with the audio speeches of 24 actors, the audio song dataset is also utilized for the better training of the proposed model. The performance of all three algorithms is compared and evaluated by using the confusion matrix and accuracy plots. The empirical results show that the KNN classification model of ML has performed better than other proposed algorithms with increased accuracy of 76.25%.