audio CDs and cdrdao vs. cdparanoia

[Urs Thuermann <urs@isnogud.escape.de>]

When I rip audio CDs, I typically use both cdrdao and cdparanoia and
compare the results to make sure that I really really have the correct
digital audio data.  I run Debian testing with current versions of
cdrdao 1.2.2 and cdparanoia III release 10.2.

For each CD I run

        cdrdao read-cd --datafile data.cdr --device /dev/sg0 toc
and
        cdparanoia -d /dev/sg0 -B

where /dev/sg0 refers to an Plextor Ultraplex 40max SCSI CDROM drive.

Then I use my own small program to split the data.cdr file into wav
files <n>.wav according to the toc file.  I then compare these wav
files with the track<n>.cdda.wav files from cdparanoia.
Alternatively, one could run

        sox track<n>.cdda.wav cdda.cdr

and then compare data.cdr to cdda.cdr.  I most cases the results of
cdrdao and cdparanoia are the same but for roughly 1 of 4 CDs one or
more tracks differ.  Sometimes this is the case for CDs with scratches
but sometimes also for CDs with no obvious scratches where both,
cdrdao and cdparanoia don't give any error message and do not seem to
have any problems ripping the CD.  I can run cdrdao and cdparanoia
repeatedly, say 10 times, and I get deterministic results, i.e. all
runs of cdrdao give the same result and all runs of cdparanoia give
the same result but the results of cdrdao and cdparanoia differ.

Now my question is where these differences come from and which results
are the correct (better) ones.  From the output to stdout I see that
cdrdao uses the Paranoia DAE library and Joerg Schilling's SCSI
library to actually read the audio CDs.  With ldd I see it is not
linked dynamically to these libraries.  So one question is, do cdrdao
and cdparanoia use different library versions?

Regards,
urs