Subject: Re: [ecasound] 2.2 problem
From: Kai Vehmanen (kai.vehmanen_AT_wakkanet.fi)
Date: Sun Feb 16 2003 - 16:31:04 EET
On Sat, 15 Feb 2003, William Goldsmith wrote:
> OK - I tracked it down. fades (both directions) on v2.2.1 break with
> durations of < 1 sec. Those work fine on v1.9
Now this was a very interesting problem. The sampling rate is one very
problematic variable for large audio applications like ecasound. It is
needed in all kinds of places (it's needed to calculate audio position and
lenght, filter coefficients, delay lines, opening audio devices, audio
files, etc, etc), but on the other hand no valid value is available at all
times.
In ecasound 2.0 (and older) I used a default of 44100. This worked quite
well, but if there actually were any bugs related to sampling rate
changes, they were difficult to find (filter frequency response slightly
wrong if srate was different from 44100, loops wrong in ewf files if
srate!=44100, delay length errors, etc, etc -> very hard to track down).
In ecasound 2.1/2.2 I decided to finally attack these bugs and set the
default to -1. Now this turned out to be an excellent way to flush out the
various small bugs still present. If some components didn't correctly
handle the -1->valid change, it would fail horribly (and there were quite
a few of these).
Now back to our current problem. It seems that although ecasound can now
handle the -1->44100 conversions (i.e. the change is correctly propagated
to all objects and subsystems that depend on srate), we lose in precision.
I've now changed the default from -1 to 384000 (8*48kHz). This solves the
precision problem, while still allows for somewhat easier bug hunting
(i.e. scaling error in filters/audio-position calculations will usually be
large as srates higher than 192kHz are not used anywhere at the moment -
and considering the physical limitations of our hearing, this situations
is not likely to change anytime soon).
Anyway, Bill, big thanks for tracking this down! This change might also
fix some other subtle bugs that are triggered by certain specific
parameter values.
PS I've put the fix to CVS.
-- http://www.eca.cx Audio software for Linux!
This archive was generated by hypermail 2b28 : Sun Feb 16 2003 - 16:31:07 EET