INTRODUCTION

Automatic discrimination of speech and music in radio recordings is important in many audio content characterization applications. To this end, we have developed a speech / music discrimination system that combines a Variable Duration Hidden Markov Model (VDHMM) with a Bayesian Network and, at a pre-processing step, employs a computationally efficient region growing technique (see "References")

The system was developed in the Signal and Image Processing Laboratory of Dept. of Informatics and Telecommunications of the University of Athens, Greece, by Theodoros Giannakopoulos, Dr. Aggelos Pikrakis and Prof. Sergios Theodoridis.

This demo version is provided only for educational purposes without any warranty. Your feedback will help us improve the system in future versions. Please note that we are not currently distributing the source code.