I've had similar experience with faulty speech recognition. I can repeat the same word the same way into the microphone 10x and it finally be accepted on the 11th time...after I've lost credit for the question. It is frustrating to say the least, but I don't want to disable the speech recognition and lose that valuable part of the learning process.
Rosetta Stone needs to fix it. It is simply too expensive to not work properly. A stopgap measure is to turn down the precision level in options to one level below normal. It now seems to be working better at accepting my spoken words.