Automatic state discovery for unstructured audio scene classification

摘要: In this paper we present a novel scheme for unstructured audio scene classification that possesses three highly desirable and powerful features: autonomy, scalability, robustness. Our is based on our recently introduced machine learning algorithm called Simultaneous Temporal And Contextual Splitting (STACS) discovers the appropriate number of states efficiently learns accurate Hidden Markov Model (HMM) parameters given data. STACS-based algorithms train HMMs up to five times faster than Baum-Welch, avoid overfitting problem commonly encountered in large state-space using Expectation Maximization (EM) methods such as achieve superior results very diverse dataset with minimal pre-processing. Furthermore, has proven be effective building real-world applications been integrated into commercial surveillance system an event detection component.