Chris Johnson
June 30th 03, 11:53 PM
In article >,
Erwin Timmerman > wrote:
> I still have to wonder about the "threshold of human hearing" they speak
> about, because with playback systems that have a volume knob, it just doesn't
> apply. If the CD is so damn soft that you can't hear a thing, turn up the
> volume. Given that fact, what should I, if I were defining the model, pick as
> the absolute threshold level?? It would just be a wild guess. Do they pick the
> threshold level relative to the loudest peak? If that were the case,
> normalizing a wave wouldn't do anything for getting better encoding. The
> threshold level would be increased just as much as the quiet waveforms get
> louder. Do they pick it relative to the RMS level of a song? If so,
> "normalizing" it your way would actually only INCREASE the level under which
> data was dropped, so it would actually hurt the encoding result. Do they pick
> it relative to the loudness of the moment? That way the result wouldn't change
> because with every low volume part the threshold would decrease as well. All
> the more reason just to duck into the source code and find out for real. Or
> maybe a message on
> http://board.mp3-tech.org/w-agora.php3?site=agora&bn=agora_mp3techorg would
> help.
FWIW, ATH is a lookup table based on Fletcher-Munson loudness
thresholds and relative to digital full scale, not to anything else.
So it's basically a hard, multiband noise gate, at a higher threshold
for very low and very high frequencies.
My take on this is pretty much that if ATH had anything to do with
it, changing the ATH level would be just as good. I hacked a version of
a LAME shared library to be able to do that- I think the functionality
eventually made it into the original encoder. I guess notlame is more of
a just-run-it project and doesn't give users that level of control, but
if it does, problem over.
Unless, of course, Myke (hey, is this Cyco Myko of Suicidal
Tendencies?) simply doesn't like over-high crest factors, which is
totally understandable. He doesn't like overly LOW crest factors either,
and has said so. I'd say the guy is just sensitive to crest factor.
Given what I already know, if I was doing mastering for him I could give
him stuff he loved every time, just by zeroing in on the crest factor he
likes best.
That's useful to know, that people do have preference about this even
if they don't know what it is they're hearing. I could have told you
that, but Myke becomes a real-world example backing me up :D
Chris Johnson
Erwin Timmerman > wrote:
> I still have to wonder about the "threshold of human hearing" they speak
> about, because with playback systems that have a volume knob, it just doesn't
> apply. If the CD is so damn soft that you can't hear a thing, turn up the
> volume. Given that fact, what should I, if I were defining the model, pick as
> the absolute threshold level?? It would just be a wild guess. Do they pick the
> threshold level relative to the loudest peak? If that were the case,
> normalizing a wave wouldn't do anything for getting better encoding. The
> threshold level would be increased just as much as the quiet waveforms get
> louder. Do they pick it relative to the RMS level of a song? If so,
> "normalizing" it your way would actually only INCREASE the level under which
> data was dropped, so it would actually hurt the encoding result. Do they pick
> it relative to the loudness of the moment? That way the result wouldn't change
> because with every low volume part the threshold would decrease as well. All
> the more reason just to duck into the source code and find out for real. Or
> maybe a message on
> http://board.mp3-tech.org/w-agora.php3?site=agora&bn=agora_mp3techorg would
> help.
FWIW, ATH is a lookup table based on Fletcher-Munson loudness
thresholds and relative to digital full scale, not to anything else.
So it's basically a hard, multiband noise gate, at a higher threshold
for very low and very high frequencies.
My take on this is pretty much that if ATH had anything to do with
it, changing the ATH level would be just as good. I hacked a version of
a LAME shared library to be able to do that- I think the functionality
eventually made it into the original encoder. I guess notlame is more of
a just-run-it project and doesn't give users that level of control, but
if it does, problem over.
Unless, of course, Myke (hey, is this Cyco Myko of Suicidal
Tendencies?) simply doesn't like over-high crest factors, which is
totally understandable. He doesn't like overly LOW crest factors either,
and has said so. I'd say the guy is just sensitive to crest factor.
Given what I already know, if I was doing mastering for him I could give
him stuff he loved every time, just by zeroing in on the crest factor he
likes best.
That's useful to know, that people do have preference about this even
if they don't know what it is they're hearing. I could have told you
that, but Myke becomes a real-world example backing me up :D
Chris Johnson