That would be the nature of that type of microphone.
Using a (USB) headset is strongly recommended for speech recognition. Using speakers is strongly recommended against.
There is no setting to "block" unrecognized commands. The "Min Unrecognized Confidence Level" option, as the documentation notes, is solely used to exclude unrecognized command log entries.
The speech recognition system is designed to recognize speech. If you provide it input, it will process it and attempt recognition on that input.
If random sounds are getting mixed into the input, that will obviously inhibit the ability to recognize speech, just like people would have trouble hearing someone speak if there was an air horn going off in their ear.