I have calculated sample sizes in the past, but never the way I am being currently asked. I have to calculate a sample size for a phone audit, so I have the population size. There really is no mean or standard deviation to plug into the equation, or paired means for that matter.
This is where the statistician has to play 20-questions with the requestor. They are doing an audit. For what purpose? What are they trying to show? Is budget a consideration? The answers to all those questions will help guide the sample size.
I wouldn't necessarily consider the population to be fixed and finite. The population size may be conceptually infinite (the audit is for the calls done for this month, but if you want to generalize, the population is the calls that could ever have been made.)
If this is an audit for QC, then looking at some of the SAS/QC tools can help (Shewhart and Pareto charts may be useful).
Without more info on intent, it is difficult to offer specifics... keep it simple, depending upon your need for high reliability/risk, a comparatively small sample size in comparison to pop goes a long way. If you can get by the attribute sampling you might reference ANSI/ASQC Z1.4-2003 or for variable sampling consider ANSI/ASQC Z1.9-2003