The first experiment of common VUI design practices will be to test the kinds of input callers prefer to use when calling an automated phone system.
This is being put together by a group at Nuance, the largest vendor of speech recognition software. We're going to evaluate a couple flavors of speech recognition, touch-tone, and mixed-mode input (where the caller can choose between speech and dialing digits).
Nuance obviously has a vested interest in making speech recognition look good, but we're designing the experiment with input from a number of industry professionals outside the company. The goal is to have the best possible experimental process, so the results (whatever they may show) will be convincing.
After Nuance has a chance to publish its own analysis we will be making much of the raw experimental data available to others in the industry so they can perform their own analysis.
So watch this space for more details as things progress. I'm hoping we'll have another experiment to announce in a few weeks, and some results in a few months.