SVM Candidates and Sparse Representation for Bird Identification

We present a description of our approach for the "Bird task Identifica- tion LifeCLEF 2014". Our approach consists of four stages: (1) a filtering stage for the filtering of audio bird recordings; (2) segmentation stage for the extraction of syllables; (3) a candidate generation based on HOG features from the syllables using SVM; and (4) a species identification using a Sparse Representation-based Classification of HOG and LBP features. Our approach ranked seventh team-wise in the challenge and showed a poor performance in the fourth stage.