SOUNDFOREST: MONITORING FOREST BIODIVERSITY USING AI-POWERED SOUND CLASSIFICATION

Abstract: For ecological preservation and environmental management, monitoring biodiversity in forest ecosystems is essential. Conventional monitoring techniques, like camera traps and direct observation, are frequently labor-intensive, time-consuming, and have a limited geographic and temporal reach. In this paper, we propose SoundForest, an AI-powered system that analyse natural ambient audio recordings to classify and track forest biodiversity. The system continuously gathers soundscape data from forest environments using edge or remote sensors. After preprocessing to eliminate background noise, the audio data is converted into time-frequency representations like Mel-Frequency Cepstral Coefficients (MFCCs) and Mel spectrograms. Deep learning models, namely Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), and Audio Spectrogram Transformers (AST) for multi-class sound classification, are trained using these representations. The models can detect changes in acoustic biodiversity over time and recognize vocalizations specific to a species. Anthropogenic sounds, such as chainsaw activity, gunshots, car engines, and human voices, are also recognized by SoundForest in order to detect wildlife and warn of unlawful logging, poaching, or unauthorized entry into protected forest areas. Even in remote locations, real-time, decentralized monitoring is made possible by the system’s support for deployment on low-power edge devices like Raspberry Pi. Supported by the system are heatmaps of biodiversity, behavioural patterns, and species distributions.

Keywords: Biodiversity Monitoring, Acoustic Sensing, Environmental Sound Classification, Deep Learning, CNN, RNN, Audio Spectrogram Transformer, Wildlife Conservation