Towards an application for detecting intruders in wildlife regions

Sound classification is a topic that has been of a major interest for the scientific community. Recently, a low complexity solution for classifying sounds in wildlife regions has been proposed. The motivation of such a classification was in detecting the intruders from these regions. In this paper we propose a different approach, one that uses Mel-frequency cepstral coefficients in a Gaussian Mixture Model framework. The tests are performed on 4 databases of 100 recordings each. The sounds of interest are represented by recordings from humans, cars, birds and animals. In order to simulate situations as close as possible to real environments, several types of noises have been considered. The new approach proves to be more robust than the previous one at the cost of increased computational complexity. Since low complexity systems are more likely to be feasible for wildlife applications, the complexity issue is discussed and a solution is proposed.