Thankfully most devices I've been working on have a touchscreen to provision the SSID/PW, or we just ask the user to put it in a text file and inject it via a thumb drive. It really doesn't have to be this hard.
I think the audio trick is a pretty elegant hack. Device agnostic and doesn't need a lot of handholding. But it also has an impact on the BOM. You need to add that Cortex M3 + microphone to sniff it out. Otherwise the whole thing could have been run off one chip like the ESP or GainSpan GS2100.
Doing the access point-to-client switcharoo is a good one too but needs a bit more instruction on the customer side, plus a way to whack the device back into setup mode when you need it.