According to Microsoft
Speech Recognition uses a unique voice profile to recognize your voice and spoken commands. As you use Speech Recognition, your voice profile gets more detailed, which should improve your computer's ability to understand you.
http://windows.microsoft.com/en-ca/windows/set-speech-recognition#1TC=windows-7When the learn box is checked, it is learning.
When the learn box is not checked it is hopefully not learning.
Whether or not a command is recognised is very obvious. Recognised text is shown in the main window along with confidence. When a command is not understood, an event is generated:
VC.NotRecognized. We hoped that the meaning of this event was self evident.
Here is what it says about the learn checkbox in the VoxCommando wiki:
Allows you to train the speech engine to recognize your voice.
VC uses the same core speech engine as Windows and the training in VC will train the same profiles that you have in Windows. It will train whichever Windows speech profile is currently active.
We recommend you leave this unchecked (off) except when you are training the system in a quiet environment and correctly saying commands. Otherwise the system may learn incorrectly.
I recommend most users just follow our advice and leave the box unchecked. Stick with windows training tools. If you have difficulty with a particular command in VoxCommando you can train that command from the tree editor by right-clicking on the command and selecting train. This feature is not documented and the ui needs some love as well but it works. It is now on my to do list. The idea is to start with a high rejection threshold and lower it until you get sucess. You can only train a single command at one time using this technique which will help to avoid false training.
[update: preliminary documentation for command-level training tool has been added to the wiki. Valid as of VC 2.1.3.8
http://voxcommando.com/mediawiki/index.php?title=Profile_Training_Tool]