[nitrogen]

I found it very interesting that (unless I misread) you are applying a time-varying gain to the whole spectrum rather than using a simple filter or EQ. Have you experimented with setting a 31-band graphic EQ to the inverse of your audiogram, or multiplying the FFT bands by a frequency-dependent gain and using IFFT? If so, does the time-windowing approach work better?

[joerick]

I would also be curious to know this. Maybe it's possible that non linearities of hearing make the time-based approach better?