Originally Posted by jrpavel
Roberto, Thanks. Glad that you have it working.
I can do a bit more on cpu usage, but nothing earth shattering. As it is, the FFT routines take more than half of processing time.
Ok. Now it is faster than any other windows convolver I've came across.
A couple of suggestions: do you really need 64k coefficients?
Well, in fact no, is more a matter of testing the limits. But I like precise ecualization in the bass range, and it means very long filters. Probably 16k will more than suffice.
Could you explain more precisely the channel mapping issue?
Yes, this is my crossover.txt:
44100 2 4 0
Nothing special, highs goes to channels 0 and 1 and lows to 2 and 3, left and right respectively.
If I change the first line to:
44100 2 4 3F
Then I have output also in channels 4 & 5, and it happens to be a copy of channels 2 & 3, but downmixed to mono (see picture).
In fact I'm not sure of this last sentence, it could be that it is channel 3 repeated in 4 & 5, but I'm sure that outputs in 4 & 5 are identical (easy to test when highs are in 2 & 3: Image comes from center spot, nothing on the sides).
Well, I hope that it becomes clear.