Birdsong Identification

A machine learning system that classifies audio recordings of bird calls and songs, built with fine-tuned Wav2Vec2 and a feature pipeline tailored for noisy field data.

Back to home View GitHub
Pileated woodpecker with waveform of its call

Project overview

Designed to help birdwatchers, researchers, and conservationists quickly identify species from audio captured in real-world environments.

  • Finetuned Wav2Vec2 for species classification
  • Audio augmentation and noise filtering for robustness
  • Custom feature extraction focused on bird vocalization patterns

Key contributions

An audio prediction pipeline with insights into model performance, dataset curation, and deployment readiness.

Transformers Audio Processing Transfer Learning Robustness

Results

Coming soon! This project is under construction as I continue to develop my skills working with audio data and models.