Well, with any detector, one would usually calibrate against known sources of different strengths, then with the known source strength and voltage develop a calibration curve relating voltage to Sieverts (radiation equivalent dose) or perhaps in this case and more appropriately grays (radiation absorbed dose).
Siverts and greys have units of energy (joules) per unit mass, and the energy deposited depends on the type of radiation which affects the interaction per unit length of path that the radiation takes through detector. Beyond the detector would be the geometric correction for distance and size of detector which limits the number of radiation particles interacting with the detector.
I presume this is for beta and/or gamma radiation?
It depends on what you want to do. As Astro pointed out already, there are several things to consider, and everything depends on the accuracy you want to reach.
The most difficult point will probably be to convert the dose received by the diode (the energy deposited per unit of mass), into a dose, received by a human exposed to the same flux. This would be ok if the PIN diode had an equivalent composition to the human body, but it isn't - it's made out of higher-Z material. As such, it will over-estimate soft gammas and under-estimate hard gammas.
There are ways to solve this, like having several diodes, certain behind a thin lead shield, and others not, so that you can have a rough estimate of the gamma-spectrum. Other problem: you will seriously under-estimate the neutron-induced dose.
Now, if you just want a rough device of the kind "beep when I have to get out of here", then all these considerations are hair-splitting. But if you want to have a good quantitative dosimeter, then you will have a lot of calibration work.
actually i'm working on a radiation monitoring application that reads information from the gamma detector, processes it and display the monitored area radiation levels in an understandable format (probably in sievert)! so i'll need conversions like those!