Blissify - analyze an MPD library and make smart playlists

Blissify is a program used to make playlists of songs that sound alike from your MPD track library, à la Spotify radio.

Under the hood, it is an MPD plugin for bliss.

Blissify needs first to analyze your music library, i.e. compute and store a series of features from your songs, extracting the tempo, timbre, loudness, etc.

After that, it is ready to make playlists: play a song to start from, run blissify playlist 30, and voilà! You have a playlist of 30 songs that sound like your first track.

Note: you need to have MPD installed to use blissify. Otherwise, you probably want to implement bliss-rs support for the audio player you use.

Installation / Usage

You'll need clang, pkg-config and ffmpeg libraries (including development headers) to install it, as well as a working Rust installation

On Debian-based systems:

apt install -y clang libavcodec-dev libavformat-dev libavutil-dev libavfilter-dev libavdevice-dev libsqlite3-dev pkg-config

On Archlinux:

pacman -S base-devel clang ffmpeg

Finally, use cargo install blissify to install it.

Note: if you are using a raspberry pi and its corresponding ffmpeg (i.e. ffmpeg -version|grep rpi gives back something), use cargo install --features=rpi blissify instead.

All the commands below read the MPD_HOST and MPD_PORT environment variables and try to reach MPD using that. You might want to change it if MPD is listening to somewhere else than 127.0.0.1:6600 (the default). It should be fully compatible with the MPD documentation.

Analyze a library

To initialize and analyze your MPD library, use

$ blissify init /path/to/mpd/root

Note that it may take several minutes (up to some hours, on very large libraries with more than for instance 20k songs) to complete.

You can further update your library by running

$ blissify update

If something goes wrong and the database enters an unstable state, you can use

$ blissify rescan

to remove the existing database and rescan all files.

If you want to see if the analysis has been successful, or simply want to see the current files in the database, you can use

$ blissify list-db

Make a playlist

Simple version

$ blissify playlist 100

This will add 100 songs similar to the song that is currently playing on MPD, starting with the closest possible.

Changing the distance metric

To make a playlist with a distance metric different than the default one (euclidean distance), which will yield different playlists, run:

$ blissify playlist --distance <distance_name> 30

distance_name is currently euclidean and cosine. Don't hesitate to experiment with this parameter if the generated playlists are not to your linking!

Make a "seeded" playlist

Instead of making a playlist with songs that are only similar to the first song, from the most similar to the least similar (the default), you can make a playlist that queues the closest song to the first song, then the closest song the second song, etc, effectively making "path" through the songs.

To try it out (it can take a bit more time to build the playlist):

$ blissify playlist --seed-song 30

Make an album playlist

You can also make a playlist of albums that sound like the current album you're listening to (more specifically, the album of the current song you're playing, regardless of whether you queued the full album or not).

To try it out:

$ blissify playlist --album-playlist 30

Make an interactive playlist

Interactive playlists start from a song, and let you choose which song should be played next among the 3 closest songs (the number of songs displayed can be set manually):

$ blissify interactive-playlist --number-choices 5

By default, it crops the current playlist to just keep the currently played song. If you want to just start from the last song and continue from there, use --continue:

$ blissify interactive-playlist --number-choices 5 --continue

Details

If you are interested about what is happening under the hood, or want to make a similar plug-in for other audio players, see bliss' doc.