Developed independently through a one-on-one school tutorial, this project implements core machine-learning components from scratch: a Java KD-Tree for efficient nearest-neighbor search over ARFF datasets, an ARFF parser, and a suite of Python scripts for audio processing, feature extraction, and classification.
The centerpiece is a speech and deception-detection pipeline for the game Liar’s Dice. Using around a hundred recorded and Praat-annotated clips, it predicts an opponent’s dice from vocal cues and acts on that prediction.