No speech recognition software is 100% accurate, you will still have to go through the text file and manually compare it to the audio track.
What I typically do if I need a text file of an audio track, like a speech or something, is to take the audio track to a court reporting service and pay them to do it. It is usually very accurate.
If you still want to look into doing it yourself, then look up "Dragon Naturally Speaking" it is generally regarded as the best speech recognition software.