Identification of Mental State Through Speech Using a Deep Learning Approach

Identification of one's feelings and attitude through speech is a powerful medium for expressing. Finding the emotional content in speech signals and identifying the emotions in speech utterances is crucial for researchers. This paper examines how well a deep learning-based model can identify speech emotions from two well-known datasets, TESS and RAVDESS. In this research work, a proper combination of frequency domain acoustic features of thirteen(13) Linear Predictive Coefficients (LPC) and Mel Frequency Cepstral Coefficients (MFCCs) are fed into a two-dimensional Convolutional Neural Network (CNN) model for classification. According to the experimental findings, the suggested method can recognize speech emotions with an average accuracy of 99% (TESS) and 73% (RAVDESS) for speaker-dependent (SD) speech.