Comparison between Support Vector Machine and Random Forest for Audio Classification

Abstract

Machine learning techniques have been in use to identify voice samples. We consider Support Vector Machines (SVM) and Random Forest (RF) and compare their performance in detecting specific audio samples from a set of samples pertaining to many subjects. In particular, we propose an algorithm that can deal with multiple features for multiple audio samples from more than one subject. This work discusses the application and other implementation issues. As a demonstration, SVM and RF models are trained using voice samples from two subjects. We found out that SVM is 100% accurate and Random Forest is 70% accurate for subject 1, whereas SVM is 60% accurate and Random Forest is 90% accurate for subject 2. Our results show that the performance of SVM and RF varies from subject to subject.