Kaggle BirdCLEF 2026 Audio Classification Data Labeling/AI Training
Audio
ClassificationParticipated in the BirdCLEF 2026 Kaggle competition, working on multi-label audio classification using the BirdCLEF dataset. Responsible for preprocessing bird song audio data, extracting features, and applying advanced classification techniques with a deep learning model. Explored data augmentation and model distillation to enrich the training dataset and improve classification accuracy. • Utilized librosa and torchaudio for audio preprocessing including RMS silent detection and Mel-spectrogram extraction. • Implemented and trained models using PyTorch, with EfficientNetV2 as the backbone for audio classification. • Applied 5-fold stratified cross-validation and AUC as the evaluation metric to ensure robust model performance. • Planned further enhancements with noisy label augmentation, model distillation, and temporal prediction heads for improved noise robustness and detection granularity.