Use the blaze function spectrum to approximately flatten the spectra
(just divide), then normalize by a constant, so that the roughly flat
spectrum has a continuum level near unity.

RAB # 2, 3 & 5 have had scattered-light and sky subtraction performed

RAB #4 is a spectrum I produced some time ago, before writing the
       scattered-light subtraction script.

RAB #6 was taken during dark time, so no sky was subtracted

Note: The S/N is roughly sqrt(counts) (i.e. the unflattened spectra can be 
      used as the variance spectrum), except below about 300 counts
      (depending on the position on the CCD) where the noise from the 
      background subtractions start to make a significant difference to 
      the total S/N.  For this reason a more complex variance model than
      the counts is required, but it only will be a problem for RAB #6,
      which doesn't have much flux anyway.