Design and Development of a Web-Based Speech Emotion Recognition System Using Machine Learning

Due to its capability to enhance intelligent human-computer interaction through being able to discern emotional information in speech, Speech Emotion Recognition (SER) has become an area of considerable research due to its importance. SER attempts to recognise the acoustic presence of human emotional expression, through more speech signal analysis. In this paper, the author explains how a Speech Emotion Recognition system can be designed and developed through machine learning and deep learning processes. Strong audio processing and extracting features including Mel-Frequency Cepstral Coefficients (MFCC) (also called spectral feature), chroma feature, spectral feature and zero-crossing rate, are employed in the proposed system to extract the emotional information efficiently. Several classification models like Support Vector Machine (SVM), Random Forest, Long Short-Term Memory (LSTM) and Convolutional Neural Network (CNN) are trained and tested using open-source speech emotion databases. Standard measures like accuracy, precision, recall and F1-score are used to assess the level of the models. It is determined in the analysis that, CNN-based model has a better performance than the others; its accuracy and the capability to generalize are both high to the dissimilar classes of emotions. The system can predict emotions in real-time with a friendly web interface that makes it applicable to real-world operations in virtual assistants, customer service applications, the analytics of a call center, and mental health monitoring.