I am developing a program where 99% of the time i can use a set Grammar in the standard Microsoft Speech Recognition to detect words being said. However in a few circumstances, i need the user to say something. This is never going to be predictable.
So is there a way to do this with the MS speech recognition?
And are there any other speech recognition API's out there (for free), that can handle non-preset words?
There is a class called DictationGrammar that allows you to recognize dictated speech.
It also supports a "dictation context" - so you can specify that the dictated text should start/end with a specific phrase.
I think that would do the trick - technically, it's still a grammar, but it seems to fit what you're looking for.
来源：https://stackoverflow.com/questions/9229187/speech-recognition-api-without-grammar-c-sharp