CNN-LSTM Deep Learning Architecture for Animal Identification Using Acoustics

DOI: https://doie.org/10.10399/JBSE.2026305789

Rohini H M, Dr. Prabhavathi S

Keywords:

Acoustic Classification, Animal Sound Recognition, Bioacoustics, CNN-LSTM, Deep Learning, Livestock Monitoring, MFCC.

Abstract:

Animal vocalizations provide valuable information regarding species identification, behavioral activities, physiological conditions, and environmental interactions. Automated analysis of animal sounds has gained considerable attention in precision livestock farming, veterinary healthcare, biodiversity monitoring, and smart agricultural systems. Conventional animal recognition approaches mainly rely on manual observation or handcrafted acoustic features, which regularly experience poor scalability and limited generalization capability. This paper presents a deep learning-based Animal Sound Recognition framework using   Mel-Frequency Cepstral Coefficients (MFCC) and a hybrid Convolutional Neural Network with Long Short-Term Memory (CNN-LSTM) architecture. The proposed framework consists of audio acquisition, preprocessing, MFCC feature extraction, CNN-based spatial feature learning, and LSTM-based temporal sequence modeling. A dataset containing 1,942 audio recordings from four animal species, namely cats, dogs, cows, and chickens, was utilized for evaluation. Proposed CNN-LSTM model achieved an overall classification accuracy of 92.45%. The developed framework provides an effective and non-invasive solution for intelligent animal monitoring applications since the obtained precision, recall, and F1-score values confirmed that the proposed model is robustness against Noise. Proposed CNN-LSTM model outperformed several existing animal sound recognition approaches.

PDF