Audio Recognition of Animals Using Optimized Deep Learning Techniques for the Conservation of Wildlife

The classification of animal sounds has emerged as a vital tool in contemporary research, offering numerous benefits for animal occurrence records, taxonomic research, and behavioral studies. However, the problem of accurately identifying animal species based on their vocalizations remains a significant challenge, particularly in real-world environments where background noise and variability in sound patterns can hinder classification accuracy. In this paper addressed this challenge by proposing a CNN-optimized approach for classifying animal sounds. In order to enhance the number of sound samples, utilized augmentation techniques to extract animal sounds from the Kaggle animal sounds dataset. The animal sounds totally 600 audio samples are used. To improve performance, this model was developed using feature extractions from the MFCC, ZCR, and Mel-Spectrogram. The seamless deployment of forest department workers is ensured by the interpretability of our model for real-world applications related to wildlife conservation and monitoring. The main goal is to successfully identify animals using auditory properties, such as tiger, leopard, elephant, and otter noises, based on their vocalizations. Additionally, The optimized CNN and LSTM for sound classification. The Optimized CNN outperformed all other models, achieving an outstanding 98.32 % training accuracy rate.