Sample Sorter 
Machine learning as a tool for making sample-based music. 


This project is a compositional tool for making sample-based "plunderphonic" music. It analyzes a library of audio samples and determines each sample's tempo, tuning, and harmonic profile. These values are used to determine pairs of samples that, when played at particular speeds, have the same tempo, have the same tuning and are harmonically coherent. Determining which samples are harmonically coherent is posed as a positive unlabeled classification problem and solved with deep learning. 

code 


Here are some small compositions made primarily using this tool to find matching samples. They are almost entirely sample based with basic audio effects (compression, EQ) applied for blending.