It's sad that the speech-to-text movement has died. It definitely has died. 5 years ago it was very high profile, IBM was bragging about Dragon, MS was bragging about this big name research group they were assembling, lots of voice recorders advertised their automatic text capabilities. Now the voice recorders don't advertise that at all, they just present themselves as audio recorders, and in fact a whole mess of listen-and-type-along software has popped up in the wake of the dying speech-to-text behemoth.
The first voice recorder I had years ago came with a light version of the IBM speech-to-text engine, and it was indeed worthless. I spent about 30 minutes reading the training text into it (you read 5 minute chunks, and it kept telling me insufficient recognition, train more), and after that it proceeded to mangle the recognition so badly as to be useless. I had thought that some errors would be no big deal, presumably they would be phonetic errors and I would still get the gist of it and be able to fix it or see what I meant, but in fact the mess that came out was so bad it took minutes of trying to figure out WTF I meant.