[vorbis] Compression Artifacts at -q 5. Help!

[Lawrence Wade]

At 02:38 AM 6/10/2003 -0400, I wrote:
>...a big long rambling question which boiled down to the fact that Red Hat 
>6.2 ships with a defective compiler. (No wonder those setiathome people 
>only provide binaries...)

    Thanks for the help, everyone.

    I was having problems not only with oggenc producing crappy files, but 
also with sox crashing when I tried to adjust the volume of a file. Indeed, 
both problems seem to be related to Red Hat 6.x's defective compiler 
because both went away when I compiled from source on another system.

    Initially, I tried to install binaries on the Red Hat machines, but it 
was getting to be too much of the usual cyclical nightmare of dependencies. 
("Can"t install vorbis-tools. Need lssl. Can't install ssl.blah.rpm, need 
curl. Can't install curl, need ssl.blah.rpm. Can't install libvorbis, need 
mod.so.impossible_to_find.2." [grumble grumble] Screw curl, it's not like 
I'm gonna try to stream oggs with a machine that has no sound card...)

    I've installed Slackware 8.1 on the bunch of old relics, and they're 
chugging away happily now. No more bad-sounding oggs, no more sox crashing. 
Everything seem to be working pretty well.

    As a user rather than a developer, I have a couple of thoughts about my 
experience, and some of the answers I got to my initial question:

    - it would probably take no more than a couple of lines in the 
configure script to have it check for borken compilers and warn the 
potential user that the output binaries will be useless. Right?

    - average users (ie. those who really think that their "300W" speakers 
powered by the 9V 300mA AC adapter are capable of putting out 300W) are 
potentially not going to notice the flaws in the ogg files created with 
defective compilers. Playing hard rock into a pair of very nice Celestion 
Ditton 44 speakers, I didn't really notice the flaw for a couple of 
minutes; I thought my ears were playing tricks on me and that I was being 
hypercritical. (Space Oddity and Pink Floyd's Comfortably Numb confirmed my 
fears that the first few CDs I'd ripped were broken.)

    - Ogg files are still pretty rare on Kazaa and other file sharing 
systems. Flawed oggs being generated by bad compilers hurt the credibility 
of the ogg as a high-fidelity alternative to the MP3. Most end-users aren't 
gonna know how or why, they're just not gonna bother anymore. (Of course, 
none of those shared oggs are gonna be copyright material...)

    - don't assume the user will always be running a new machine. Ask no 
questions whatsoever, and try to keep the software requirements as low as 
possible. Low software requirements will keep the hardware requirements 
down, which increases the portability and enhances the ability of the 
software to be used anywhere. It's also in keeping with Unix' frugal 
nature. (Ken Thompson wrote Unix on a scavenged PDP-7.) Support for old 
compilers and libraries should continue as long as it doesn't cause an 
impediment to progress.

    - Red Hat is the most popular distribution of Linux, and 6.x versions 
remain quite popular still running on workhorses all over the place - I can 
think of a dozen still in use among workplaces and friends. While it's 
reprehensible that their compiler was defective, it's a fact of life.

    - Finding a modern but really compact no-frills text-only 
takes-no-time-to-install lemme-ping-another-host-and-compile-libvorbis 
Linux distribution is getting tough.

    - When one attempts to rip a scratched CD using cdparanoia, one can 
also oggenc another file on the same host. Even if the host is a mere 
486DX-66, oggenc will finish encoding a CD before cdparanoia has finished 
ripping the next one. This allows two instances of crip to run concurrently 
on all 6 hosts, processing 12 compact discs at any given moment. The 
486DX-33 lacks cache RAM and so is painfully slow at encoding, doing one CD 
every 14 hours or so. (Usual copyright personal use only disclaimer applies...)

    - As an extension of the really great crip utility, I toyed with the 
idea of creating a live CD which would boot, create a swap file and /tmp in 
empty space on a hard disk's existing filesystem, request an IP address by 
DHCP, set itself up as a Samba server using the machine's MAC address as a 
hostname, and allow passwordless incoming telnet sessions for control, then 
eject the boot CD. The system would then accept a CD, crip it under control 
from the telnet session, and make the files available to be removed. A 
quick and dirty distro like this would be convenient to have a collection 
of machines easily and painlessly encoding and provide Windows users an 
instantly useful Linux introduction.

    - Ogg kicks butt. I used to work in the professional audio field and 
have done sound for The Three Tenors, Harry Belafonte and Garth Brooks, to 
do a little namedropping. At -q 5, I cannot consistently tell the 
difference from the ogg and the original wav file. With MP3, I cannot do 
that until I've encoded with the Fraunhofer encoder above 192kbps, which 
results in a considerably larger mp3 than the ogg.

    - Rudimentary consistency and sanity checking as suggested by one of 
the replies would be a great idea if it can be implemented reasonably 
efficiently. Of course, given the fact that this is necessarily a lossy 
codec, this would be imperfect but might serve as the canary in the coal 
mine before a user floods a filesharing system with bad oggs of his high 
school choir. Can it be done effectively enough to catch fidelity problems 
more subtle than mine? It's hard to predict the next bad compiler or 
whether the Pentium V will have the same math error as the Pentium 60.

    Thoughts? Flames? "You've completely missed the point!" rants?

    Thank you all for Ogg Vorbis and the help,

    Lawrence Wade

<p>--- >8 ----
List archives:  http://www.xiph.org/archives/
Ogg project homepage: http://www.xiph.org/ogg/
To unsubscribe from this list, send a message to 'vorbis-request at xiph.org'
containing only the word 'unsubscribe' in the body.  No subject is needed.
Unsubscribe messages sent to the list will be ignored/filtered.