You can find the new version of the PiFm radio transmitter at http://www.icrobotics.co.uk/wiki/index.php/Turning_the_Raspberry_Pi_Into...
It now uses just 5% of the cpu instead of 100%. It allows you to pipe audio into it from other programs. It uses DMA, and has an audio buffer for constant rate glitch free playback.
Finally, it also implements FM pre-emphesis, which gives the resulting sound the correct amount of "bass" and noise immunity.
The FM pre-emphesis filter is described elsewhere on the web, but nobody has presented an algorithm to generate such a signal. Various questions have been asked on forums, yet gone unanswered. MATLAB can't generate such a filter for you, because the magnitude response tends to infinity as the frequency tends to infinity, and hence the filter wouldn't work.
In the real world, the FM signal is bandlimited, and it turns out the implementation of such a filter is easy:
out[n] = in[n] + (in[n-1]-in[n])/(1-tau*samplerate)
That also boils down to a nice two tap FIR filter!