Comparison study on speech emotion prediction using machine learning

Abstract

Speech Emotion recognition is the process of analyzing human emotions. People vary widely in their precision at recognizing the emotion of others. Support vector machine (SVM) is one of the best algorithms for prediction and handling the voice data. In this speech emotion identification is a difficult task that includes various preprocessing steps. The proposed one, the method is based on the Cepstral Mel frequency Coefficients (MFCC) and voice signal energy as Input and use of the RAVASS emotional database function discourse. In this research work support vector machine algorithm, linear kernel processing is used to predict the human emotions.

In Feature Extraction using Librosa module to handling audio and music type of data. Librosa can extract the values of Mfcc-Mel-frequency Cepstral coefficients it’s used for identifying monosyllabic words in voice recognition. Chroma mainly used for similarity and audio matching. In voice data each data are applied to feature extraction technique. Then generated matrix format of values for calculation. In proposed Support vector machine algorithm for act as a classifier to find emotions. Mfcc, chroma, Mel features are directly send out the Support vector machine (SVM) algorithm. Its take a class of eight emotions like happy, sad, calm, angry, neutral, fearful, disgust. Based on our features (mfcc, chroma, Mel) to classify emotions. The comparison study on SVM and KNN algorithms, SVM has predicted 4% better accuracy than KNN.