That's great, but can't you use the subaudible tones as an additional blocking metric? Or even as an assessment metric to flag potential false negatives?