Needs to be more than 50% accurate. Each show has one host or two hosts and sometimes there are multiple guests on the program. The program should adjust and improve transcriptions over time...similar to dragon. Each radio show would have it's own "account" or run a separate copy of the program so it would adapt and improve its transcriptions based on that show. It would learn to choose the most common voice for the shows on that account or program copy, and adapt it's transcribing based on that, while recognizing other voices/dialects and either transcribing them based on a generic voice...or have the program recognize and merge similar voices/accents/dialects and transcribe and adapt based on the closest store voice it has. Ideally, there would not be a training or setup period.