Using recorded audio from just about any source, including an iPhone, Scribe transforms a wide variety of audio file types into text, but oddly enough it won't work with .mp3 files. This isn't a major problem since you can convert them to .aac files in iTunes, but not inviting .mp3 files to the party seems like a major oversight.
After playing a few minutes of audio into Scribe, you train a small portion of the transcription by correcting mistakes in a few paragraphs. This trains the program to recognize the voice. Scribe can handle six unique voice profiles. Once done, Scribe takes in the audio at about twice the speed of speech and delivers your text file in a raw format, but with no punctuation. If you included words like 'comma' and 'period,' they'll be turned into actual punctuation. Using the Nuance engine, it's remarkably good. I've found MacSpeech Dictate to be around 95%, and MacSpeech Scribe should be just as good since it's using the same engine.
I can see this being incredibly valuable for students who want to record lectures, and have a printed copy. But there are many other uses. I did my share of ethnographic research which entailed many hours of recorded interviews and transcribing the data. This was both tedious and mind-numbing. MacSpeech Scribe does the grunt work, since fixing punctuation is a snap when compared to transcribing dozens of hours of speech. I would have killed for something like this in grad school.