An Application Programming Interface to Recognize Emotion using Speech Features

With the recent increasing interest in the study of the emotional component of speech signals, a number of methods have been put forth to ascertain the emotional content of uttered words. Often, we can tell how someone is feeling by looking at their faces. Another method is to detect someone’s emotions by auditory cues or speech. With a more straightforward architecture and fewer learnable parameters, the aim of this paper is to identify someone’s emotion from an audio input. Three aspects have been suggested in the proposed methodology named MFCC (Mel-frequency Cepstral Coefficients), Mel-spectrogram and Chromagram, and MLP (Multilayer Perceptron). These features have been employed in this paper, because they perform better. MLP is a supervised machine learning, which is frequently used in various research to categorize human voice recognition. In addition, MLPs are appropriate for classification prediction issues where inputs are given as a class or label. They are also appropriate for regression prediction issues in which a real-valued quantity is forecasted from a collection of inputs. With the help of our proposed methods, we are able to achieve an accuracy of 80% on the RAVDESS Dataset.