Olympus-OM
[Top] [All Lists]

Re: [OM] (OT & OT) Panorama problem

Subject: Re: [OM] (OT & OT) Panorama problem
From: Ken Norton <ken@xxxxxxxxxxx>
Date: Tue, 19 May 2009 10:52:55 -0500
>
> See http://www.dynamicrange.de/ (in english)
>


This reminds me of my own contribution to the volume wars.  (sorry).  Back
in the '90s I was teaching recording and mastering engineers how to build
multi-band compressors in digital editors and how to achieve a maximum
dynamic range of about 2dB. We could get it to sound great, but the fatigue
level was over-the-top.

Today, there are essentially two systems used to master the audio. The
mastering process is essentially taking a 2-channel mix (sometimes 8)
provided by the recording facilities and creating the final editions for
Album, Broadcast and Dance.  Each has its own specific needs. The two most
commonly used compressor systems (available as either a stand-alone 2RU
rackmount box or as a software plug-in) were developed by companies I worked
closely with. One took a more conservative approach and the other the
over-the-top approach. Unfortunately, one of them so thouroughly destroyed
the audio that you could do absolutely NOTHING to it afterwords. Radio
stations also compress and limit their audio. The songs mastered on the one
system sounded so horrid on air that the stations had to alter the
compression settings to provide levelling and limiting only. This caused
these stations to lose their own signature sound!

The other system, which I was personally involved with the development of,
is now in its fourth major iteration and provides up to five distinct bands
of compression per channel with smart adaption to the material. It can
distinguish between types of sounds within the bands and also determines
rythm.  Extremely advanced stuff--all designed to give the abilty to squish
meters up against the peak without wiggling.  I personally use Version 2,
but have a self-created processing-matrix to give me three bands of
compression. (I'm doing that which didn't make it in production systems at
the time)

What all of these idiot recording and mastering engineers don't understand
is how bit-rate-reduction algorithms like MP3 work.  All MPEG-based BRR
schemes require audio information to be "discarded". They don't throw away
sounds, but what they throw away is the detail (bit depth) of those sounds.
Although,...  All algorithms use a "look ahead" feature which plays on the
physiology of the human hearing. When confonted with a loud sound, the human
hearing actually discards the sound immediately proceeding it.  For argument
sake, when you are listening to a "proper" jazz recording and the drummer
has a distinctly loud kick or snare hit, you actually do not "hear" the
proceeding 50ms of music. The algorithm knows this and will take that
proceeding 50ms of audio and will discard the detail of most of it. Also,
the algorithm will take the next 100ms (or so, as it is somewhat adaptive)
and discard much of the detail of that too.  Furthermore, if you have two
notes playing, say the middle-C on the piano, and a softer playing middle-D
on a flute, the algorithm knows that the piano's note will "mask" the
flute's note. Therefore it will keep the detail of the piano and will
discard most of the details of the flute.

Inotherwords, MPEG-based algorithms have to have audio information to
discard.  Wide-band noise, like white or pink noise is not tolerated by MPEG
and the algorithm MUST discard something to maintain the required output
bit-rate.  Each MPEG sampling block will have to throw away something and
each MPEG sampling block is calculated seperately from the proceeding
sampling block. (however, there are adaptations which average multiple
sampling blocks, but this will backfire on some types of audio).  Dolby
compression is slightly different because it uses a fixed algorithm on every
block and is not adaptive. As a result, you can cascade Dolby compression
without further losses and it retains time-coherency for surround-sound
purposes.

I say all this, because certain instruments, like a cymbal are "wide-band"
noise generators. When passed through MP3, the algorithm has to discard
something and each sampling block is essentially independant from the
previous block (with some limits). The end result is a cymbal sound which
sounds like it is being flushed down the toilet. Wide-band sounds will
"swirl".

When you hyper-compress (dynamically squish) the audio with these fancy
mastering compressors, you create an audio signal which the
bit-rate-reduction algorithms can't handle. There isn't anything to
discard!  Therefore, everything starts to swirl!

Once you really get tuned into the artifacts, you'll be pining for old vinyl
recordings.

There is a variation of the mpeg-based algorithms that utilize the limited
dynamic range of the mastered audio and will automatically alter itself from
a standard 16-bit depth to 2-4 bit depth. It then applies an "offset" value
for the final output level. So, if you have a song that has a fade in and
fade out, what it will do is use only 2-4 bits per sample (before further
analysis and discarding), but applies the offset that says that the volume
is down, say 20dB. Therefore it fakes a greater bit-depth than is actually
there. I actually co-wrote the first version of this back in 1994 but it
never got approved by the consortium and we didn't incorporate it either as
it appeared at the time that 24-bit depth was going to be the future and we
put the development dollars into 96/24 instead.

AG
-- 
_________________________________________________________________
Options: http://lists.thomasclausen.net/mailman/listinfo/olympus
Archives: http://lists.thomasclausen.net/mailman/private/olympus/
Themed Olympus Photo Exhibition: http://www.tope.nl/

<Prev in Thread] Current Thread [Next in Thread>
Sponsored by Tako
Impressum | Datenschutz