Sound event classification using bidirectional local binary pattern

Abstract

Sound event classification is a basic approach for real-world application. This paper presents feature extraction technique for classifying audio data. The audios are classified by combining of digital signal and digital image processing. Firstly, the audio data is partitioned into fixed length and each portion is transformed into spectrogram image. The, the distinct features are extracted from this spectrogram using bidirectional local binary pattern. Finally, support vector machine and k-NN are used for classification. The method is tested on an audio database of ESC-10 sound event data.