Thorough enough to make the person on the other side of the phone think they were talking to the intended gender. The point of the experiment is to control for gender, not other factors like style of speech. That would need a separate test.