There is currently no explicit option to set the listening state on startup, however there are a few ways to achieve that effect:
You can start VoiceAttack with the "-listeningoff" command line option (you can add this to the shortcut you use to start VoiceAttack, assuming it's a shortcut directly to the executable)
You can configure a command at the profile level, that runs when that profile is loaded as the active profile, using the "Execute a command each time this profile is loaded" option, which would disable listening
You can configure a command at the profile level, that runs when that profile is unloaded while it was the active profile, using the "Execute a command each time this profile is unloaded" option, which would disable listening
In case of either command, you'd want to either check the previous profile using the "{PREVIOUSPROFILE}" token, or the next profile using the "{NEXTPROFILE}" token, respectively, to check whether VoiceAttack is starting up or shutting down, so the listening state isn't changed each time the profile is loaded (though note that if you reload the profile without having first switched to another profile, "{PREVIOUSPROFILE}" will return blank as well)
Also note that this only applies to a single profile. You'd need to either set this up in all your profiles, or have the command run at startup while setting that profile to load on startup using the "Load Profile on Startup" option on the "General" tab of the VoiceAttack options window.
Lastly, if you set the "Recognition Global Hotkey" option to either "VoiceAttack listens while keys are down", or "Listen then stop listening after recognition completes", VoiceAttack should start with listening suspended.
Note that if you have this option set to "VoiceAttack stops listening while buttons are down", the inverse effect should apply, and may interfere with the other methods listed.