Augmented Speech Emotion Recognition Using Machine Learning

Speech emotion recognition (SER) is a technology used to analyze emotions from spoken language. These systems aim to classify a speaker's emotional state by analyzing subtle vocal characteristics like pitch, tone, and rhythm. However, background noise and distractions can significantly reduce the accuracy of emotion detection by SER systems. The Wiener filter acts as a digital noise-canceling tool, specifically targeting speech signals corrupted by environmental noise. Mel- frequency cepstral coefficients (MFCCs) enhance the speech features. The work revolves on increasing the feature count followed by exercising feature selection using the filter method. Select- K best provides the stage for the application of four chosen machine learning (ML) models namely SVM, Random Forest, Decision Tree, and KNN. Random Forest outperforms other models in its performance. The work is an elementary analysis on ML to explore the need of other high end DL architectures like CNN, RNN and LSTM for SER. Work is in progress to port the most accurate algorithm to an embedded platform like Raspberry Pi for integration with IoT.