Linux

Hah, pipewire, sorry.

I was describing my setup from memory, here's what I actually have:

play "`find <path-to-clips> -type f -name "*.wav" | rofi -dmenu -i -fuzzy`"

I then have this bound to a key combo in my qtile config which will pop up a menu so I can fuzzy search for a clip to play. However, this alone will just play the clip. To get it to go through discord, I also have a file ~/.config/pipewire/pipewire.conf.d/50-null-sinks.conf with this in it:

context.objects = [
    {   
	factory = adapter
        args = {
            factory.name     = support.null-audio-sink
            node.name        = "mic-loopback"
            node.description = "Mic Loopback"
            media.class      = Audio/Sink
            audio.position   = [ FL FR ]
            adapter.auto-port-config = {
                mode = dsp
                monitor = true
                position = preserve
            }
        }
    },
    {   
	factory = adapter
        args = {
            factory.name     = support.null-audio-sink
            node.name        = "input-mix"
            node.description = "Input Mix"
            media.class      = Audio/Sink
            audio.position   = [ FL FR ]
            adapter.auto-port-config = {
                mode = dsp
                monitor = true
                position = preserve
            }
        }
    }
]

This creates two new nodes that I can use to combine audio into a single source. I then use a combination of the pavucontrol and qpwgraph GUIs to control what audio streams go where. I wire my actual microphone along with the output ports of "Mic Loopback" to the input ports of "Input Mix". Then whatever app I want to play back through my mic, I wire up to the input ports of "Mic Loopback". To wire SoX up to Mic Loopback, I play a clip that's long enough for me to make the switch, and then it tends to remember that for the next time SoX launches. Finally, I wire Input Mix up to discord or whatever program I'm using.

The rofi/play combo is rock solid, I really like that. The pipewire/qpwgraph/pavucontrol part could probably be improved. It can feel a little non-deterministic, but really I think I don't fully understand what each app involved is doing to manipulate the pipewire graph, or how to configure them so they don't try to override each other when a pw node is added/removed.

Edit: quick note. You might be wondering, what's the point of having both Mic Loopback and Input Mix? Couldn't I just have applications go directly to Input Mix? Yes, but generally you want to also hear the sound yourself, without hearing your own mic, and you want to be able to pick a single output device from within the app (most apps don't let you choose more than one). Having two separate nodes lets you split off the output of Mic Loopback to both Input Mix and the output associated with your headphones/speakers, that way you don't hear yourself too.