Using Spectrogram To Detect North Atlantic Right Whale Calls From Audio Recordings

this work describes a number of experiments aiming to assess the use of spectrogram to detect North Atlantic right whale calls. For this purpose, spectrograms are generated from the audio signal and then, Local Binary Pattern textural features are taken from the spectrogram images, once texture is the main visual content found in this kind of image. Following, these features are sent to a Support Vector Machine classifier. The experiments were carried out taking into account different ranges of variation of the amplitude looking for the best setting to highlight the textural content of the spectrograms. In addition, some different values of SVM classifier parameters were evaluated. The best accuracy rate obtained is about 85.48%.