Thanks for that. I had already ruled that explanation out because looking at the frequency output with a meter verified that is was actually really a problem, not merely a perceived one.
Also, using recorded audio won't work in my application because I need highly accurate tones lasting 200mS, 500mS, 800mS, and 1S with seemless transitions from one to the next.
Thankfully, I have an alternative method for doing this now.
Thanks again for the response.