I dropped the idea because I first want to increase the efficiency of the computation by using the Altivec unit of the Macintosh G4 which allows to compute 4 floating multiplications and additions in one single instruction. Also I'll try to approximate fonctions like 'sinh' or 'log' by a polynomial expansion.

I have previously implemented the Mathematical interface between the high-level control parameters of a Parametric EQ and the low-level filter coefficients.

Here is what a Parametric EQ looks like in the frequency domain.

The user can change up to 5 parameters :

• F0: the peak gain/cut frequency (Hz)
• Bw: the bandwidth specified in octaves
• G0: the initial gain at DC (dB)
• G: the peak gain/cut (dB)
• GB: the gain at bandwidth edges (dB)

Here is the Math that leads to the filter coefficients :

This computation is quite expensive. My project is to find a model for these Mathematics to reduce the amount of computation and allow a much greater number of EQs in cascade. I am currently thinking of using Neural Networks or Cluster-Weighted Modeling.