Arabic Speech Emotion Recognition Method Based On LPC And PPSD

Abstract

This research detects and recognize the emotions in Arabic speech audio files that contains records of human voices with different emotion classes (sad, happy, surprised, and questioning). In the area of emotion detection, when a person becomes emotional, his voice is adjusted based on the state of emotion. As the acoustic features like pressure, strength and loudness varies from a state of emotion to another. However, in the detection of feelings, the classification and modeling part of the features gets priority with the extracted features. Therefore, extracting the best features that describes the emotions stats is the most challenging task. This paper proposes an efficient approach to recognize the Arabic speech emotions. The presented method contains three main phases, signal preprocessing phase for noise removal and signal bandwidth reduction, feature extraction phase using a combination of Linear Predictive Codes (LPC) and the 10-degree polynomial Curve fitting Coefficients over the periodogram power spectral density function of the speech signal and machine learning phase using various machine learning algorithms (ANN, KNN, SVM, Decision Tree, Logistic Regression) and compare between their accuracy results to get the best accuracy.