I briefly considered going/staying fixed point for processors, but I didn't see a real benefit. Obviously, you need to take a coefficient like 0.93 and scale it up by some factor of 2^n or whatever so it can exist as an integer. And then there's more scaling down in the end. So, it seemed that all the fussing around might negate any benefit. In the end, you only have so many bits to work with. I was happy with what I was hearing with 64 float so I left it as is.
Interesting test!
On Fri 23/11/12 7:47 AM , Fons Adriaensen fons@linuxaudio.org sent:
On Fri, Nov 23, 2012 at 07:01:09AM -0600, grekimj@acousticrefuge.com wrote:
> There are a couple of outstanding (commercial) fixed point filters
> out there, by the way.
Yes, but the question is: are they outstanding because they are
fixed point, or because they have been designed well generally,
with the fixed point aspect just being marketing ?
A few years ago there was a very large scale listening test
(involving hundreds of 'trained' listeners), comparing the
quality of 'HD' DVD-A and SACD recordings reproduced directly,
and the same recordings reduced to CD quality. There's an
engineering report about it in the AES library.
The first result was that essentially nobody was able to
identify which was which.
The second result was that most listeners actually preferred
the HD recordings to CD versions of the same music, same
performers, but not he same recording.
That seems to be a contradiction, but it isn't. The simple
fact is that those HD recordings, targeting a specialist
audience, have been made with much more care and expense
than the average CD. They are actually better recordings,
but *not* because they are 'HD'.
I'm pretty sure something similar is involved with e.g.
the fixed point filters you refer to, and a lot of similar
things. Just going for quality doesn't sell things, because
most people won't hear the difference. But adding a bit of
pseudo-science helps, in particular if a large part of the
target audience has no technical or scientific background
anyway and is not capable of separting facts and fiction.
Ciao,
--
FA
A world of exhaustive, reliable metadata would be an utopia.
It's also a pipe-dream, founded on self-delusion, nerd hubris
and hysterically inflated market opportunities. (Cory Doctorow)