I decided to enter the BirdCLEF+ 2025 competition on Kaggle, and it was the most revealing experience ever.

Our Task

The primary objective of this competition was to classify animal sounds as belonging to an animal from a provided list. The training data had $29,000$ examples of labeled data. The data was audio recordings re-sampled at $32$ kHz.

My Approach

Situation

Prior to this project I had very little experience working with audio data. To the best of my knowledge, the only audio processing project that I’ve worked on worth mentioning was for a synthetic music generator that a friend asked me to build. Other than that- nothing.

Task

Action

To address this issue, I formulated a comprehensive plan-of-study for to address my knowledge gap, and also expose myself to the adjacent domains that would theoretically help me do well on this project. I chose to select the topics most relevant to the project task. I didn’t know what I didn’t know, so a lot of decision had to be made qualitatively (a good ol’ vibe check).

Digital Signal Processing (DSP)

Digital Siganl Processing is the fundamental [piece of science; ] that I used to work on this project.