For all the hardware combinations I have tried so far, Windows has detected the device as an audio device and so there has been no problem selecting that device in the OS. I have not tested them all with Julius on Windows, however.

In Linux, the experience is this: all the headsets I have tried with the exception of the Bluetooth headsets have been recognized by the system as audio devices. In Opensuse Yast, they can be configured and tested and allocated a priority as a sound device. Not so the bluetooth adapter/dongle and headsets. This is an issue for Julius since issuing

$ julius -input mic -C ....

results in Julius offering a <please speak> but never hearing anything from bluetooth.

Fortunately Pulseaudio sees the headsets as audio devices even if the OS does not, and

$ padsp julius -input mic -C ....

works. (Thanks to the Julius forum for this bit of info.)

However it is a bit more complicated. When detecting with wireless Julius sometimes spits out a bunch of warnings related to stripping which can be eliminated with

$ padsp julius -nostrip -input mic -C ....

(thanks to another post on this forum for this tip)

There is still another issue which I have not quite resolved with wireless, involving Julius and warnings about "no data fragment"

These two sets of warnings are a nuisance in that they can interfere with Julius' detection of speech with false positives and false negatives, and for sure your dialog manager needs to be able to deal with them.