[linux-audio-user] Advice needed re: latency tuning/optimization, esp. video

davidrclark at earthlink.net davidrclark at earthlink.net
Wed Feb 18 09:52:26 EST 2004

At the risk of confusing the issue even more or providing more misinformation:

There is a lot of misunderstanding out there about the Live! cards, partly
due to the lack of information from Creative Labs on the chipset.  The
internal clock runs at a fixed rate: 48 Ksamples/sec.  All digital data
are at this rate for all processes.  This means that, for a recording session:

The analog signal is sampled at 48 Ksamples/sec.  If you store it at
44.1 Ksamples/sec, it is downsampled.  When you play it back, it is 
read from the storage medium, upsampled to 48 Ksamples/sec, the converted
to analog.  This is why, IMO, it sounds so bad.  It has been resampled
twice.  On the other hand, the card is capable of performing much better
than this by working at 48 Ksamples/sec instead of 44.1 Ksample/sec.
(48,000 is popular due to the large number of prime factors of two;
but 44,100 is 2^2 * 3^2 * 5^2 * 7^2 so also has its advantages.)

The Terratec EWX-2496, for example, is not fixed.  The internal clock
runs at whatever rate you specify.  If you specify 44.1 Ksamples/sec,
then this is what it runs at.  Analog signals are sampled at 44.1 Ksamples/sec,
then stored at this rate (assuming you've set it up that way).  When
they are played back, they are read at 44.1 Ksamples/sec, then converted back
to analog.  This results in more accurate sound --- what you hear on
playback is more like what you heard when you did your recording.
(I myself work at 24/96, mixing and all that, then downsample as
the last step.  I do this because I do a LOT of processing.)

Now it is an interesting side effect that if you burn a CD, the Live!
mixed result will sound better on another stereo system (one that runs
at 44.1 Ksamples/sec) than it did when you were doing your mixing.  Has
anyone else noticed that?  Suddenly things have improved.  Well, that's
to be expected because the signal was not upsampled.  Although some people
may be happy with this miraculous result, I myself am dissatisfied with this
way of working because it never sounds like it what I heard previously.
Playback for mixing sounds worse than what I heard during recording, yet
the CD sounds better than the final mix.

In summary, the Live! series of cards are providing you with very misleading
information about the sounds you are working with if you insist on working
at 44.1 Ksamples/sec.  You'd probably be better off working at 48 Ksamples/sec,
then downsampling.  A major problem, though, is that many sample libraries
are at 44.1 Ksamples/sec.  In this case, you are forced into upsampling
for audition and mixing!  So...  It all depends on what it is exactly
that you are doing.  This is definitely not a "one size fits all" situation.
I wouldn't begin to tell anyone what they should do --- just provide some
information about what is happening so they can make their own decisions.
Do be careful!!

As a footnote, Steve Harris mentioned that the Live! cards resample, even
at 48 Ksamples/sec.  Well, the signal path may include resampling, but
technically, the interpolation algorithms used should reproduce the input
exactly.  So in fact, there should be no difference.  In other words, don't
be misled into thinking that resampling is occurring anyway, so you might
as well work at 44.1 Ksamples/sec.  No, you're worse off, assuming that
there are no side issues such as heavy use of 44.1 Ksample/sec samples.
Once again, it depends on exactly what it is that you are doing.

I do have a Live! Value card myself, but avoid these issues by refusing to
use it for serious audio work.

More information about the linux-audio-user mailing list