GitHub - TOMAL5085/BirdCLEF-2026: The goal of this competition is to develop machine learning frameworks capable of identifying understudied species within continuous audio data from Brazil's Pantanal wetlands. Successful solutions will help advance biodiversity monitoring in the last wild places on Earth.

🏆 Competitions

Bird Sound Classification Challenge

Built an end-to-end audio classification pipeline to identify 234 bird species from soundscape recordings.

Converted raw audio to Mel Spectrograms (128 bands, 32kHz sample rate) and fed them into an EfficientNet-B0 model using the timm library with ImageNet pretrained weights.
Implemented custom PyTorch Dataset and DataLoader pipeline with audio augmentation (random gain adjustment, Gaussian noise injection) to improve model generalisation.
Performed comprehensive Exploratory Data Analysis (EDA): class imbalance analysis across 234 species, audio quality rating filtering, and multi-source data integration (iNaturalist + soundscape recordings).
Designed the full inference pipeline: loaded trained model weights, ran sliding-window chunk predictions over soundscape files, and generated final submission using sigmoid-based probability thresholds.

Tech Stack: Python, PyTorch, timm, EfficientNet-B0, Librosa, OpenCV, NumPy, Pandas, Matplotlib, Seaborn