Two methods come to mind. The first is, using a PIC or AVR, to simply measure the pulse duration, cross reference this through a calibration table to an output frequency, then output that new frequency from an output port.
The other method is to use convert the pulse duration to a DC voltage, then use this DC voltage to drive a voltage variable multivibrator. Don't have any parts in mind for this one.
I infer from this that for every 1 mph increase the freq. increases 2 Hz and the duty cycle doesn't matter, within reason. The speedometer can then resolve 1/2 mph.
This is probably the simplest way to convert rotation into frequency. The speedometer is then a freq. meter. Going to a PW circuit seems like an extra step.
You could check it by running the sensor output into a one-shot adjusted for a pulse width of ~ 1 mS and see if the speedometer still tracks the speed up to 60 mph.
From experience, this test circuitry needs to be shielded from RFI from ignition noise or horns blowing.