Spatiotemporal Acoustic Manifold

Each bird species produces a unique acoustic fingerprint. This project extracts 57 audio features per frame from birdsong recordings — including MFCCs, chroma, spectral centroid, and onset strength — then uses PCA to reduce that high-dimensional space down to 3 principal components. The result is an interactive 3D trajectory rendered in real-time with Three.js, animated and synced to the audio. Colour maps to energy (blue → silence, orange → peak call). Size maps to amplitude.

What I Built

A Python audio processing pipeline (librosa + scikit-learn) to extract, scale, and reduce audio features to 3D
A Three.js web app with a live comet-trail animation, orbit controls, and audio sync
A binary search for frame-to-time mapping for smooth real-time playback

Tech Stack

Technology	Role
Python	Core language
librosa	Audio feature extraction
scikit-learn (PCA)	Dimensionality reduction
NumPy	Numerical computing
Three.js	3D rendering
WebGL	Graphics
ES Modules	JavaScript modules

Skills Demonstrated

Machine Learning
Audio DSP
Data Visualisation
3D Graphics
Python pipeline design

Why This Project Matters

The Three.js + PCA combination is a strong differentiator — it shows you can bridge the ML pipeline and the frontend output end-to-end.

This project demonstrates the ability to take raw audio data, apply machine learning techniques for dimensionality reduction, and create an engaging visual experience that brings the data to life.