I've spent a lot of time searching for a solution to this, and I think my lack of experience with speech server development is what's holding me back. I've designed a basic UCMA 2.0 application that I can successfully call/debug from my OCS 2007 R2 deployment. Unfortunately, I struggle with on thing: the default voice is very bad, and not something I'd consider usable in an application that will face either my or my client's customers. How do I get better, natural-sounding voices out of OCS 2007 R2 / UCMA 2.0?