Worf

Worf is a daemon that automatically queues songs in MPD based on a "pin", which is just whatever song was playing when the daemon was started. It uses bliss-audio to queue songs that are most similar to the pin. The pin is changed simply by playing a new song.

Usage

First, initialize the bliss library if not already done by running worf --base-path $MPD_BASE_PATH init in the project root, where $MPD_BASE_PATH is where music is stored (can be a network location with username/password). Once that's done, run worf daemon in the project root to start queueing similar songs. Use worf update to update the bliss library with new songs from MPD.

By genres (experimental!)

Credit to Glenn McDonald and Spotify for the Every Noise at Once project, which provides similarity metrics for Spotify's genre tags. If your music is tagged accordingly, such as with Zotify (or my zotify-genre-tagger if you forgot to enable genre tagging), worf can queue music by genre similarity. While this prevents the sort of "drifting" that purely audio-based similarity metrics might cause, in my experience, it often leads to the opposite problem of staying too close in a genre bubble.

You can try it with worf genres. You will need a genres.json file in the same directory, with genre names as the keys and a 5-element array of integers as the values. I won't post it directly here, but you can acquire your own copy with this Javascript on everynoise.com:

let weights = {};
for (let node of Array.from(document.querySelectorAll(".canvas .genre"))) {
    weights[node.innerText] = [parseInt(node.style.top), parseInt(node.style.left), parseInt(node.style.color.split(",")[0].split("rgb(")[1]), parseInt(node.style.color.split(",")[1]), parseInt(node.style.color.split(",")[2].split(")")[0])]
}
JSON.stringify(weights);