So far, the two fastest alternatives that I've seen are speech recognition and machine stenography (see http://www.openstenoproject.org/), both of which seem to be designed for typing valid English words, not the kinds of complex systems of strangely-spelled, hard-to-pronounce, and symbol-heavy text used in code and documentation.

I don't actually have a problem with the keyboard, but typing out individual characters seems antiquated when better alternatives must exist.