
Feature Extraction and Machine Learning (II)
A class session from MIT's Computational Music Theory and Analysis course, taught by Michael Scott Asato Cuthbert. Guest speaker Kiki Gutierres opens with his computational research on Pinkfong's Baby Shark song and related musical corpora, showing how large sets of recordings can be mined for structural patterns. The class then turns to feature extraction methods used in music information retrieval, working through how melodies get converted into numerical representations that machine learning systems can compare. Students and instructor analyze similarity and difference across melody examples, discussing which extracted features actually capture meaningful musical distinctions versus noise. The session sits late in the course sequence, assuming familiarity with earlier machine learning and music theory material, and runs as a working class discussion rather than a polished standalone talk, with screen-shared code and data examples driving much of the analysis.