RogerBW's Blog

I have a multi-channel home audio setup that works quite remarkably well.

And all I wanted was the ability to say "I've finished being here for a bit, so when you get to the end of the current track, pause playback". Then I got interested.

The core of all this is a complete separation of roles: a control client, that determines what's being played, is completely separate from a storage server, which has access to the files, and a playback client, which actually wiggles a speaker cone – even if they end up running on the same machines. This requires a central audio server, and a host for each set of speakers or headphones – a desktop or laptop computer can do this as part of its other jobs, or for a stand-alone player a Raspberry Pi will do the job nicely. (Snapcast even comes with a custom lightweight OS image that just runs the Snapcast client and nothing else.)

This setup comes in two parts: mpd, the Music Player Daemon, and snapcast. Both of these have packages in Debian/stable, which is what I run; if you're using something else, see MPD homepage and Snapcast homepage.

MPD first. Configure this to point at your audio repository. Note that all filenames must be valid UTF-8 to be indexed. (Also, it'll read cue files, but this is generally an error unless you're storing pairs of cue/flac; when I get music that way, I prefer to break it down to individual file-per-track.)

If you want another channel, i.e. a separately playable mpd instance that shares the same music database as the primary instance, set up a new config file, changing playlist_directory, log_file, pid_file state_file and port to new values – and start it separately. Also set:

database {
	plugin "proxy"
	host "localhost"
}

MPD can drive speakers directly, or serve http, but in this case we'll be running the output through snapcast. Here's how I do that:

audio_output {
    type        "fifo"
    encoder     "flac"
    name        "snapserver"
    format      "44100:16:2"
	path		"/tmp/snapfifo"
	compression	"8"
	mixer_type	"software"
}

Other mpd instances get a different path: /tmp/snapfifoB etc.

Then add a matching entry for each instance in /etc/snapserver.conf:

source = pipe:///tmp/snapfifo?name=mpd-A&sampleformat=44100:16:2&codec=flac
source = pipe:///tmp/snapfifoB?name=mpd-B&sampleformat=44100:16:2&codec=flac

etc.

Each of those is a stream name from mpd: mpd-A, mpd-B, etc. (and in theory you could link in other streaming sources too, but I don't).

Then individual snapclient installations can be pointed at the server, and configured either with the official web interface or my command-line tool (which I wrote because I didn't feel I needed yet another thing driven through a web browser).

The end result of all this is that I have multiple players, all synchronised, so that if I'm on my own in the house I can have audio from everywhere – and of course the API in mpd makes it very easy to control playlists and such like. (See other "MPD" posts.)

mpd-tools package on github

snapcast-tools package on github