Abstract: In this paper, the system submitted by the UVigo-GTM for the Albayzin 2010 Audio Segmentation Evaluation is described. The task is performed in two different stages, a segmentation stage and a classification stage. The segmentation stage employs an approach called BIC-POISSON, which consists in a modification of the BIC algorithm oriented to reduce false alarm rate in the acoustic change detection stage. The classification is performed with Support Vector Machines, using the segments obtained in the segmentation stage.
Index Terms: audio segmentation, BIC, poisson, GMM-UBM, SVM.