Abstract: This paper presents a method for audio segmentation that separates broadcast news audio files into five acoustic classes for Albayzín Audio Segmentation 2010. The proposed system makes use of a presegmentation stage based on the Bayesian Information Criterion (BIC), a music/speech classifier based on a combination of GMMs and a binary decision tree, and finally a speech/speech with music/speech with noise classifier based on GMMs.
Index Terms: Audio segmentation, Bayesian Information Criterion, Gaussian Mixture Models, C4.5 Tree.