Playing with synthetic data in python

I'm exploring/researching which combination of FSK frequencies and symbol length would be best for demodulation and in particular for correctness of the sync word detected position (I'm for now generating synthetic sampled data in python).

I've read in video circles the audio sampling frequency is often 48 kHz and I want the sync word to be the cleanest possible so I opted for 00 as the binary values and a space frequency to be an exact divisor of 48 kH, ie 1 kH for space and 2.1 kHz for mark. Here is the word 101011 following the 00 sync word:

 

 (Kinks at transitions will be filtered out by a low pass analog filter on the board). Also for the sync word to be easily detected, it should be composed of an exact multiple of fmin period (1 kHz), here a bit takes 4 periods of fmin, ie 4 mS. Experiment will show if this could be shortened to 3 without losing precision in post-prod.