Explore the creation of automatic song remixes using audio signal processing and machine learning in this 30-minute video tutorial. Learn about the Infinite Remixer Python application, which generates remixes by patching together multiple songs at similar beats using beat tracking and Nearest Neighbours search. Dive into the system's code, design rationale, and usage instructions. Discover experiments conducted with the system, including the use of chromograms and MFCCs, as well as adjusting the "jump rate." Examine the shortcomings of the current implementation and potential improvements. Gain insights into linear and non-linear music consumption, and understand the concept behind projects like The Eternal Jukebox. Access the Infinite Remixer GitHub repository and explore additional resources on music psychology and expectation.
Automatic Song Remixes With Audio Signal Processing and Simple Machine Learning
Valerio Velardo - The Sound of AI via YouTube
Overview
Syllabus
Intro
Linear music consumption
Non-linear music consumption
The Eternal Jukebox
First look at Infinite Remixer
Segmentation component
Data component
Search component
Nearest Neighbours search
Remix component
How to use Infinite Remixer
Experiments with Infinite Remixer
Using chromograms
Using MFCCs
Experimenting with the "jump rate"
Problems with the system + possible improvements
Outro + project GitHub
Extended remix example
Taught by
Valerio Velardo - The Sound of AI
