Subject: Re: [linux-audio-dev] Random thought on HDR latency compensation
From: Tom Pincince (stillone_AT_snowcrest.net)
Date: Wed Apr 26 2000 - 01:51:48 EEST
Thanks folks, I have plenty to think about now. I will go back to
lurking for a while, but before I go I have one more thing on my mind.
My 15 years of contemplation started the first time I heard an audio cd,
in the fall of 1984. I thought it sounded terrible. My mind
immediately started to search for answers to the question, "what is
wrong with this sound?" In 1985 I shifted my focus to composing and
performing electronic meditation music, and this is still my main
focus. In the process I have learned to silence all intellectual
processes while still maintaining consciousness, and have spent much of
the last 15 years in this state. The question, "what is wrong with this
sound?" did not go away, I simply saw answers without thinking about
them. Aside from a/d/a issues that are not relevant to this mailing
list, the main problem that I saw was the compounding of rounding errors
caused by applying multiple mathematical processes to a single sound
file. Since EQ, pan and fader controls found on most hd mixers all
introduce their own rounding errors, these errors add to the
pre-existing quantization error from the a/d to produce a compound
quantization error that extends well past the least significant bit.
Plugins add to this problem. I read the thread that focussed on double
precision float for plugins that addresses this issue somewhat. The
method that came to my mind is to treat a 24 bit sample as a 32 bit
signed integer with 5 of the extra bits added to the left for an
additional 30db of headroom, allowing 32 soundfiles hitting 0dbfs
simultaneously to be mixed without producing overs, and the remaining 3
bits added to the right to accumulate the rounding errors which would
ultimately be removed when the audio is reduced back to 24 bits for
final output. Because I saw this approach in meditation as opposed to
directly attempting to implement it, I have no idea what the practical
issues are regarding the choice of integer vs floating point math. Any
high level (plain english) explanation of this, or direction to
pre-existing documentation on this subject would be greatly appreciated.
Tom
This archive was generated by hypermail 2b28 : Wed Apr 26 2000 - 02:29:58 EEST