It crashes a lot.
I’ve been using Soundbooth to clean up speech from about 10 people recorded in a conference room, on a tiny digital recorder, while it sat next to a projector’s fan, under an air-conditioning outlet, in a slightly echoey conference room, open to a common area where people gather to take the elevator.
Cleaning up the audio was a reasonably fast, if crashy, experience.
Once I had a reasonably clean recording I thought I’d try out the automatic transcription feature. This feature performs some cursory speech recognition on an audio file to produce tags that match recognised words to their location in the audio file along with the speaker’s identity (e.g. speaker 2). As it’s only 75% accurate (for my samples, forgivably, it was much worse) it’s more useful for finding points of interest in an audio file than as an auto-transcription service.
I don’t think it ever completed the transcription task. At some random point it would offer to send Adobe a copy of a crash report.
Though it doesn’t have transcription capabilities, I think I prefer Audacity.