medical report dictation), you can achieve almost 100% recognition accuracy. If you really put some effort into it, you can learn how to speak to the system in order to minimize the word error rate. It's true that you can quite quickly dictate a short work email to your colleague asking him to provide you with some documents, or text a message with a lunch appointment, but dictating more difficult and creative texts isn't possible, or at least convenient. The performance of the major commercial ASR systems (I don't want to name any of them here) is quite impressive considering the situation 10 years ago, but I still think that their advertisement and commercial presentations are somewhat exaggerating.

So instead of building own ASR from an open-source toolkit (which could be very difficult), it's probably much more efficient to purchase a commercial system for 100-200USD. The good ones are often not cheap, but I think they're affordable. There are ready-to-use applications, but definitely not free.

Open source speech to text software