Hi Norman et al,
That's definitely a cool service. The folks at the Center for Civic Media have even used it to build tools to make it possible to interface with websites via cell phone (see: VOIP Drupal).
For the purposes of hyperaudio, however I'd warn that the quality just isn't there yet for anything that needs accuracy. It might catch some percent of what you say, but that percent is far lower than you need for a transcript!
There are services out there like SpeakerText and 3Play which use a hybrid Machine Learning + humans (Mechanical Turk) to try to reap the best of both worlds and even then quality remains a potential issue. Maybe in 10 years we'll get there...
On a slightly different note, has anyone seen tools or services out there that can take an existing transcript and time sync it? That is a much easier problem from what I have heard, but I haven't seen anything that does it yet. Maybe we should go implement it :D
Best,
- Dan