BirdNET Model V2.4

BirdNET model version 2.4 is an advanced deep learning model designed for avian sound identification, capable of recognizing over 6,000 bird species globally. It features an EfficientNetB0-like backbone with a final embedding size of 1,024, optimized for performance and model size. The architecture includes a custom spectrogram layer, followed by 2D convolution, pooling, the EfficientNetB0-like backbone, global average pooling, and a linear classifier. This version also incorporates an enhanced species range model that utilizes eBird checklist frequency data to estimate species occurrence probabilities based on location and time, improving identification accuracy.