Well, I don't know why you are getting the messages you are getting, but have you tried setting up an initrd (initial RAM disk?) I found that having Speakup load from an initrd allowed me to shut it up immediately as it comes up later in the boot process. From what I've read about kernel 2.6, the major difference from 2.4 is that 2.4 and earlier assumed anybody who has a computer has a keyboard. 2.6 makes no such assumption, and thus, could in theory, work with an embedded system which can only be accessed remotely. I think what's happening is that keyboard support comes up later in 2.6, and thus, you can't stop speech until it detects the keyboard. As I said, with Speakup built entirely as modules, and loading it from an initrd, it comes up a little later, but you can silence it immediately. It's a compromise between compiling it directly in the kernel, and having to wait until you are fully booted up to start speech.
