Thank you Lee!
No doubt, the accuracy of grammar and pitch will be low. However, I do not think it is of high importance in this case. Just getting the words down into a text format would allow for some interesting data mining given enough surveys (and verbatim).
I was looking at DNS as well, but it mainly seems like a dictating tool for personal use. Nuance seems to have a broad range of services to offer though, including CC tools.
Some research led me to CMU Sphinx, which seems very potent for the right (wizzy) person.
The hunting continues :)