I guess there are several ways to skin this particular cat. However, it seems to me to be a waste not to use the inherently most accurate aspect of the input here: the fact that the O.P. has a source of pulses that are, more or less, exactly equal to the speed (rpm = frequency). Therefore, as suggested by alec_t, why not just count those pulses? All that's needed is a minimal amount of signal conditioning (squaring). The only real problem I see is the fact that since the tach output voltage is also (more or less) proportional to speed, some measurement at the bottom end will be lost as the voltage approaches zero. (This would presumably also be a problem with F-V conversion.)