Full spectrum with Lame? Options

[sld]

I seriously think, however, that we need to find out _why_ this has happened with Lame.

This has happened with LAME to make it the single most consistent and reliable mp3 encoder in the 1st decade of the 21st century. There were actual brains and ears doing research tweaking LAME to such a level.

Oh well, if it can't be done then that's that. Thanks for all of your help everyone.
David

There's still older versions on ReallyRareWares. I think I'll rather you pick a LAME version from there than use any version of Blade's.

This post has been edited by sld: Feb 22 2009, 21:52

[MichaelW]

Like I said, I can't ABX between the two. But I can certainly see the difference. One thing I learned when getting into science was that we can't trust our bodies (I think that's the basis of perceptual encoding), so we use machines to display the information to us differently.

Trying to be non-confrontational: AFAIK, it is true that we can't trust our bodies for absolute measurements, and this is precisely the principle behind perceptual lossy encoding. With our ears, we cannot tell very well whether the highest frequencies are kept or dropped. If the purpose is to analyse the spectrum of a complex signal, we can't rely on our ears. But, conversely, if the purpose is to produce a good-sounding signal, economically, then spectral analysis doesn't tell us much about how our hearing will react, and it is mistaken to believe that a particular spectral response is important without, or in opposition to, empirical evidence.

So it's a question of purpose. Lossy encoders are designed for signals that sound good. It seems, from what I read here, that the good lossy encoders have been cutting down on the high frequencies, for good reasons. This might be a disadvantage if you want to preserve the high frequencies for some good reason: there are some people who really can hear the difference, and one of those might be in the intended audience; but that doesn't seem to be the case here. Or the lossy encodes might be wanted for some reason other than listening to; but in that case, a perceptual lossy encoder is pretty certainly a bad choice, anyway. Lossless, or LossyWAV, would be better.

So what you're wanting doesn't fly in the face of any religion, but is in contradiction to the fundamental design principle of LAME.

[gerwen]

So what you're wanting doesn't fly in the face of any religion, but is in contradiction to the fundamental design principle of LAME.

Quoted for truth, and this is the crux of the argument.

Lame is designed to sound as good as possible, discarding what you'd like to see, because you can't hear it.

You're picking an encoder based on what is aesthetically pleasing to the eye, rather than picking one that is pleasing to the ear.

The 'religion' around here, is to use a scientific method (abx) to find what is most pleasing to the ear in the least amount of bits, which aligns nicely with what Lame is trying to do.

[TJA]

Great

Still such an similar topics - after all that years.




May i hint at my signature:

3.90.3 --alt-preset extreme -V0 --lowpass 20.5 -> yeah!
"extremist of extreme", johnV @ Sep 13 2002 - 02:01 PM ;-)

3.90.3 is old, but still good enough for me.
The "--lowpass 20.5" is overkill - but read further:

I recommend reading all those old topics like the following - and *numerous* others!
http://www.hydrogenaudio.org/forums/index....f=17&t=4173

And of course:
http://wiki.hydrogenaudio.org/index.php?title=Lowpass

So, even if 20.5 may be hearable, only 20 could make sense at all - and in practice, 19 may be the maximum perceivable at all.
So, add "--lowpass 19" for a try

BUT: All this is honored in the standard-settings, depending on the quality setting and resulting file size:

http://wiki.hydrogenaudio.org/index.php?title=LAME

Scroll down to "Technical information":

CODE

Switch      Preset      Target Kbps      Y Switch      Lowpass      Resample
-V 0     --preset fast extreme     ~245         19383 Hz - 19916 Hz     
-V 1         ~225         18671 Hz - 19205 Hz     
-V 2     --preset fast standard     ~190         18671 Hz - 19205 Hz     
-V 3         ~175     Y     17960 Hz - 18494 Hz     
-V 4     --preset fast medium     ~165     Y     17249 Hz - 17782 Hz     
-V 5         ~130     Y     16538 Hz - 17071 Hz     
-V 6         ~115     Y     15115 Hz - 15648 Hz     
-V 7         ~100     Y     14581 Hz - 14968 Hz     32000 Hz
-V 8         ~85     Y     12516 Hz - 12903 Hz     32000 Hz
-V 9         ~65     Y     9336 Hz - 9602 Hz     24000 Hz

So, just use "-V 0" on 3.98.x or "--vbr-new -V 0" on 3.97.x or my setting above for older versions.

Hough!

This post has been edited by TJA: Feb 24 2009, 18:35

[timcupery]

I understand why people are giving davecool crap. I do understand where he's coming from - thinking that the lower lowpass might be perceptibly lacking in some sort of environment, even if typical listening tests don't capture that.
But then, the issue is to find such environments and run ABX tests in them, and my guess is that these haven't shown up anything.

So the point that LAME's lowpass has changed over the years is interesting to me.
I don't see this as a problem (I trust that LAME has been developed carefully); it's basically an interesting comment on how much we've learned about where high frequencies aren't detectable by the human ear.

This post has been edited by timcupery: Feb 24 2009, 19:18

[Garf]

So the point that LAME's lowpass has changed over the years is interesting to me.
I don't see this as a problem (I trust that LAME has been developed carefully); it's basically an interesting comment on how much we've learned about where high frequencies aren't detectable by the human ear.

It's not so much about believing where high frequencies are detectable or not. It's a compromise between putting the lowpass too low, which risks being audible in itself, or putting it too high, which risks wasting bits on inaudible frequencies only to screw up the lower part of spectrum.

[timcupery]

It's not so much about believing where high frequencies are detectable or not. It's a compromise between putting the lowpass too low, which risks being audible in itself, or putting it too high, which risks wasting bits on inaudible frequencies only to screw up the lower part of spectrum.

Understood and agreed. Not sure how/if this runs counter to what I wrote.