[sdiy] Re: [Fpga-synth] Hello! and an Idea!

[Magnus Danielson]

From: "JH." <jhaible at debitel.net>
Subject: Re: [sdiy] Re: [Fpga-synth] Hello! and an Idea!
Date: Wed, 6 Sep 2006 11:48:34 +0200
Message-ID: <000f01c6d199$9f8bf970$0200a8c0 at jhsilent>

> >Well, just to be a bit nasty to your nice little discussion... using FFT on 
> >a
> >transient signal (and transient responce) is actually making a gross
> >oversimplification of the signal and transient responses and hence the
> >artifacts.
> 
> I think that performing any Fourier-like transformation on a slice of
> audio signal "asumes" that the same material is repeated infiniely (cyclic),
> and creates the Spectrum accordingly. (What else could it do, not knowing
> what's outside that slice?)
> But I _thought_ with making the slices twice as long, giving 50% overlap
> with both, preceding and following slices, and applying a window function,
> allows to get rid of these artefacts. Or so I've heard. I have no idea if
> a perfect emulation of time domain convolusion is possible by that, or just
> an approximation; and neither do I know how much extra processing power
> is needed to achieve this. So just take it as hearsay from my side; but I
> think I read it's done that way in commercial convolution boxes.

You never get rid of them, but you can certainly combat them to become less
apparent. The most annoying aspect is the unexpected jump between the first and
the last sample. You want some sort of cross-fading between ending and
beginning slices, unless you run so few slices as 2 where you have to suffice
with pure window-functions to zero-out in the ends.

Cheers,
Magnus