Niels Elemans wrote on Sat, 24 April 2004 05:24 |
Could someone explain to me why a good compressor algorithm must be very DSP/CPU intensive? Of course you have to process in a really big word length to keep soundquality after gainreduction. Or has it something to do with complex attack/release curves? And does a higher sample rate matter for making a accurate algorithm, I heard it does matter for EQs.
|
Well, close. Yes, we want to use higher sampling rates to get the control path right.
As I mentioned, most processors are naturally linear; even floating point instructions are built from linear, simple algebraic operations. What we understand of perceived dynamics have algorithms much more to do with exponential (logarithmic) relationships. These are expensive to craft, mips-wise.
Also, as much as I love and use products based on convolution techniques, I would advise you not to be charmed by manufacturer's claims for modeling dynamics. They are not telling you the truth - it does not work. Yet.
George