Thanks for your bug report. This has always been the case unfortunately. The Windows speech recognition system is a heavy resource to set up and can take anything from one to many seconds to initiate. It depends entirely on Windows and whether its speech recognition system is in a busy state. I would guess that it is the type of feature that is meant to be enabled at the startup of an application and only occasionally updated, instead of being repeatedly enabled/disabled/updated with a script. There is little mIRC can do to speed up this process I'm afraid.
