The ability to recognize emotions in speech has the potential to enhance a variety of areas, including safety, customer service, mental health, and communication. Speech-based emotion detection focuses on classifying audio recordings according to specific emotions. The frequency and pitch are read from the audio files. We are contrasting the accuracy of the two models, the Multi-layer Perceptron Classifier model and the Support Vector Classification model, in order to attain the goal of recognizing the fundamental emotions, such as calm, happy, sad, and angry. We used the Ryerson Audio-Visual Database of Emotional Speech and Song (RAVDESS) dataset with MEL-Frequency Cepstral Coefficients, Chroma, and MEL as feature extraction techniques. The outcomes demonstrate that the MLP classifier outperformed the SVC model. For the MLP and SVC models, 92.34% and 70.69% accuracy scores, respectively, have been recorded.