Hivanta-Animal Voice Recognition Using Machine Learning

The field of bioacoustics has witnessed significant advancements through machine learning, enabling automated species identification. This paper presents Hivanta, a web-based animal voice recognition system that classifies animal sounds using a Convolutional Neural Network. The system processes audio inputs via Mel-frequency cepstral coefficients and delivers real-time predictions through a TensorFlow.js-powered interface. Evaluated on a dataset of $5,000+$ samples across five species, the model achieved 92.6% accuracy with a prediction latency of $<1.5$ seconds. This paper demonstrates the viability of lightweight, browser-based solutions for ecological research and education, addressing limitations of manual sound analysis.