If you're referring to speech as user input... I think it has some limited usage, especially to the disabled. However, most people have the preconception that the next big step in UI will end up being voice recognition for everything.
The problem is that voice input is inherently much slower, and far less accurate. The gains it offers are usually only in dictation, and often a good typist can obliterate any time savings there. For application usage, there would need to be some excessive advances in speech recognition in order to offer any competition for the precision and speed that normal keyboard and mouse input already delivers.
The real places where voice input can shine is applications where a keyboard and mouse input is inconvenient-- PDA and smartphone type products. The problem here is portability, and it's hard to get efficient and low power performance to portables that's sufficient to do the work that analyzing voice requires. Specific chips dedicated to voice recognition might solve some of these problems, but this doesn't seem like a very elegant solution.
I think, ultimately, that the biggest area for research and development lies with solving UI problems for portable devices. These kinds of technologies can carry over into the desktop world and ultimately offer the biggest progressions in using computers. UI paradigms only come along so often...
