This would be an improvement over other transcription interfaces I've seen because you normally seem to have to listen through from the beginning to get to the part you want to hear, which is time consuming, since you can't skim audio the way you can skim text. (Jott is good, but designed for short messages, and I am imagining something designed for long-form content.) The ready availability of the transcription process would make it actually useful to record things like class lectures, or to keep an audio diary, because you can search and skim through the material; or if you want to hear a particular passage, you can select as much text as you want to hear and issue a Play command.
Maybe the interface could be like OneNote and let you build up a reference library, by maintaining time/date metadata for each item and letting you add tags (aside -- am I weird to want hierarchical tags? or put another way, multi-folder categorization.)