[fruityloops] Re: Nominal Sound Forge Rendering . . . (continued)

  • From: "Gwydion Elderwyn" <Gwydion@xxxxxxxxxxxx>
  • To: <fruityloops@xxxxxxxxxxxxx>
  • Date: Thu, 18 Jul 2002 18:21:32 +1000

I just noticed that FL uses 6-point hermite.  Same principle, even more
accurate, also slower again to calculate.

The "sinc" methods should be used with care, however.  Whereas linear and
hermite retain the original points (ie, the line or curve passes through the
original points), sinc interpolation actually replaces everything with
calculated points.  Sinc interpolation attempts to reconstruct the audio
signal by using a series of overlapping sine waves to produce what's called
a "best fit" curve.

Upside of sinc interpolation: supposedly, it's the most "musically pleasing"
of all of them because it creates the smoothest possible audio curve.
Downside: very slow to calculate (hence the various depths, 64 is faster
than 128 and 256 is the slowest of all), and as I said it actually changes
the original signal, so what you heard from FL when it was playing will be
slightly different to the rendered result.  Is it audibly different?  Don't
know, never done any comparison tests.  Might make an interesting project,
though :)

One last note: bearing in mind that most of what we do ends up as mp3 as
well, if you're going to use a good interpolation method, make sure you use
the best mp3 compressor you can get as well.  I use LAME on the highest (and
slowest) quality setting and even at 128k I have a hard time telling it from
my original WAV file.


Other related posts: