Wild Animal Recognition: A Vocal Analysis

The Animal Voice Recognition system has become an important application in the field of animal sound retrieval and classification; it is highly beneficial for applications involving bio-acoustics and audio retrieval. Currently existing work lacks wild animal vocal dataset and requires appropriate vocal feature engineering mechanism even in the presence of varied background noise is necessary to conduct vocal analysis for efficient recognition. In this research work, initially wild animal vocal database was created, then fusion of vocal features were extracted for developing the animal recognition system based on Multiple vocal features such as the Mel-Frequency Cepstral Coefficients (MFCC), Zero-Cross-Rate (ZCR), Spectral Roll Off, Spectral Centroid and Spectral Contrast features. Further, PCA is used to extract the transformed discriminating features. Finally, the performance of this model has been analyzed by extracting features of the collected dataset using number of conventional classifiers to identify the specific animal. Among which SVM and Random Forest classifier obtained 96.89% and 97.46% recognition accuracy respectively and it proves to be cost-effective.