Splitting a song into its individual instrument, vocal, and other audio tracks—for example, to create a backing track for a favorite song—usually requires expensive AI software. Not so with the surprisingly polished StemKit on GitHub, which is released under the MIT license. Here's how it works.