A Reliable speech emotion recognition framework for multi-regional languages using optimized light gradient boosting machine classifier

In today's world, the interpretation of emotions from human speech has prompted a lot of research attention in signal processing applications. Many speech recognition approaches have been introduced to detect the psychological states of individuals for public datasets. However, identifying the different emotions in the Indian regional language is one of the most demanding promises. As a consequence of inappropriate feature selection techniques, existing machine learning classifiers produce more misclassification errors for Indian regional datasets. Therefore, this work proposes a novel machine learning TEL (TF-EPOA + LGBM) classifier for accurately predicting human emotions. In the first stage, the pre-processing of the speech samples can be carried out to eradicate the artifacts. To express personal emotions in speech, the threshold-based feature selection technique is utilized to select the appropriate speech features. The optimal speech features are fed into the Light Gradient Boosting Machine (LGBM) classifier to significantly determine the emotions. The hyperparameters of the LGBM classifier are properly tuned with the aid of a Hybrid optimization where the Tangent Flight Operator (TF) is incorporated with the Exponential Pelican Optimization Algorithm (EPOA). This effective tuning facilitates the proposed classifier to achieve a superior classification accuracy of 99.27 %, 98.4 %, and 96.7 % for the Tamil, Malayalam, and English language datasets, respectively.