作者: Jozef Vavrek , Eva Vozarikova , Matus Pleva , Jozef Juhar
关键词:
摘要: Audio classification is one of the most important task in content-based analysis and can be implemented many audio applications, such as indexing retrieving. This paper addresses problem broadcast news classification, by support vector machine - binary tree (SVM-BT) architecture, into five classes: pure speech, speech with music, environment sound, music sound. One substantial step creating architecture selection an optimal feature set for each SVM classifier. Therefore we implement F-score algorithm, effective search within a space characteristic features that mostly used speech/non-speech discrimination.