[gentoo-amd64] Re: libmad sound artifacts

[Duncan]

"=?UTF-8?Q?Canek_Pel=C3=A1ez_Vald=C3=A9s?=" <caneko@×××××.com> posted
ee2a916a0803261624o7f4d264ag4aed324956007d69@××××××××××.com, excerpted
below, on  Wed, 26 Mar 2008 17:24:10 -0600:

> Finally, I was able to locate the problem: it's libmad, because madplay
> have the artifacts too.
>
> Everything that uses libmad sounds like crap, and I've recompiled the
> library with any option that it offers, but to no avail: the artifacts
> are still there (I can make them WORSE, with the default fixed point
> math). I see bug reports that *seems* to be similar, but for ppc and
> sparc, and those solutions didn't work for me.
>
> Other thing: if I decode the MP3 to a WAV file with madplay
> --output=wave:file.wav, the artifacts disappear.

Interesting, as I think wav is basic uncompressed PCM the same as what 
would go direct to the speakers anyway.  The difference then, isn't in 
output, it's that the to-wav-file doesn't have to be processed as real-
time, while obviously sending it to the speakers does.

The strong indication is therefore that the problem isn't in the 
processing per se, but that the libmad processing is either too CPU 
intensive for your system config in real-time, or that it's defaulting to 
an incompatible sampling rate for your system and/or soundcard, and 
sending it to wav first uses a different process and/or sample rate.

From what you said, it /used/ to work fine.  Assuming that's the case, 
the problem may not be with libmad at all, but with recent kernel changes 
and/or latency or priority changes on your system.  It's a pretty big 
field of possibilities to try to narrow down from, IOW.

Or, it could be the sampling rate, as I mentioned.  I don't know, does 
libmad allow fiddling with that?  When was your latest upgrade and does 
the changelog mention anything about new sampling defaults or the like?  
(FWIW, here I'm no libmad-0.15.1b-r4, with the dates on the files 
suggesting I last merged it on March 17, which would have been the emerge 
-e world after I upgraded to gcc-4.3.0.  I am however on ~amd64 with gcc 
and binutils always unmasked to the absolute latest in-tree regardless of 
keywording, and always emerge --upgrade --deep.  If you are on stable and/
or don't regularly use --deep, you may have an earlier version; epkginfo 
says -r2 is latest stable, but 0.15.1b is going to be the latest upstream 
release.)

One thing to note is that libmad isn't the only mp3 decoder available.  
ffmpeg (IIRC) handles mp3 as well, and I believe there's actually around 
a half dozen different implementations.  As such, especially where a 
package has USE=mad instead of USE=mp3 controlling its dependency on 
libmad, you likely have either a compile-time or run/config-time choice 
as to its mp3 decoder.  Thus, it's quite likely you can simply use 
something other than libmad, and go about your business, altho it may 
involve recompiling a few packages.  Try equery depends libmad and see 
what shows up and what USE flag controls that dependency, for starters, 
keeping in mind that general file conversion isn't affected so letting 
say k3b and streamripper, to name two apps I have installed with mad 
support here, continue to use libmad isn't a problem.  xine-lib, OTOH...

> I'm not using normal speakers;
> I'm using an old stereo system, plugging the output from my sound card
> to the video input of the stereo. But it sounds really nice if I'm not
> using libmad.

The following has nothing to do with your problem, and is just general 
comments on using the stereo as a computer sound output device...

I have my stereo system hooked to my computer too.  Way better sound that 
way, IMO.  Only I have a relatively new Onkyo 5.1 system, and run coax 
pcm/digital-out computer -> Onkyo.  I had it hooked up analog b4, but 
prefer the digital, except now the sound card volume doesn't work, only 
(on the apps that have it) individual application volume, on the 
computer, and of course the stereo's volume.  Someday I need to find 
documentation on getting the multi-channel simulation on the computer to 
work, as ideally, it should give me more control and options, but for 
now, my sources are almost all simple stereo anyway, so I just send that 
to the Onko over the digital/PCM interface and use its 5.1 simulation.

-- 
Duncan - List replies preferred.   No HTML msgs.
"Every nonfree program has a lord, a master --
and if you use the program, he is your master."  Richard Stallman

-- 
gentoo-amd64@l.g.o mailing list