GitHub - iamsrp/playmake: A simple playlist generator which uses audio similarity to walk the song space

1 min read Original article ↗

PlayMake

Overview

A simple playlist generator which uses audio similarity to walk the song space.

Description

PlayMake takes a standard m3u playlist and puts all the songs therein into an embedding space. Once it has these, it walks the space, finding random close (Euclidean distance) neighbouring songs, appending them to a playlist as it does so.

For example:

playmake.py -G -p $HOME/Music/mp3/Pop/Pop.m3u -s 'The_The/45_RPM_-_The_Singles_Of_The_The/13-December_Sunlight_(Cried_Out).mp3'

It's not fast (even if you have a decent GPU), it is memory hungry, and it's also far from perfect. But it kinda works.

Dependencies

PlayMake requires:

  • argh for argument parsing.
  • mutagen for reading song metadata.
  • openl3 to generate the embeddings in the song space.