Abstract—An intelligent nocturnal animal identification system is proposed in this work to assist in the recognition of different kinds of nocturnal animals. Firstly, each sound sample underwent the steps of the noise reduction and the syllable segmentation. Next, the segmentation was converted into Mel Frequencies Cepstral Coefficients (MFCCs) as the main identification feature of the proposed work. To increase the recognition accuracy and lower the computation overhead, the system can guide a user to input environmental information as the parameters of the decision tree classification module. It then proceeds with the clustering, and the results are classified with a composite classifier. A series of simulations were conducted to demonstrate the feasibility of the proposed algorithm. Drawing from 18 species, 408 sound samples, and 4477 feature segmentations were utilized to verify the accuracy rate of our work. Our simulations showed that the proposed approach can effectively increase the recognition rate and decrease the training time. In particular, the collected sound sample of the Apus affinis is only one that results in a lower accuracy rate. We expect to expand the database of sound samples, and thus we expect to be able to dramatically improve on the above-mentioned problem, thereby increasing the accuracy rate for each species.

Index Terms—Composite classifier, decision trees, feature extraction, machine learning, nocturnal animal vocalizations, noise reduction, syllable segmentation, support vector machines.

The authors are with the National Dong Hwa University, Taiwan (e-mail: cjhuang@mail.ndhu.edu.tw).

Cite:Heng-Ming Chen, Chenn-Jung Huang, You-Jia Chen, Chao-Yi Chen, and Sheng-Yuan Chien, "An Intelligent Nocturnal Animal Vocalization Recognition System," International Journal of Computer and Communication Engineering vol. 4, no. 1, pp. 39-45, 2015.