Full spectrum with Lame? Options

[davecool]

Hi all, I have been trying to figure this out for a while. Maybe there's no answer...

A few years ago when I encoded most of my CDs using Lame, it would encode practically the entire audio spectrum, up and over 20 kHz. That was back when I was using 160 kbps encoding. More recently, when I use the newest version of Lame (complied from source code on OS X 10.5.6), I can't get it to encode anything above 16 kHz reliably. I typically encode at 256 kbps CBR with -h (q2).

I know a lot of people here don't care what the spectrum looks like, and only care if it sounds good. Well, yes, they still sound good, but I feel like I am missing out on this upper spectrum, and I really want it to be encoded. Plus, I always have the spectrogram iTunes visualizer running and I don't want the top third of the screen to just be black.

In any event, I have tried almost everything to get Lame to encode more of the upper-end. I have set the lowpass filter to 22050, but that didn't change it. I don't think this has anything to do with the pre-filtering. I think Lame is just reducing bits at the upper end to accommodate more at the lower-end. That makes sense to me, but when using such a high bitrate, there should be plenty of bits to go around. Even encoding at 320 kbit CBR, there is still degradation of the high end. The only way I was able to clean up that area was by doing 320 CBR with q0. That looked clean up to around 20 kHz but then completely blank above it. Even that coverage up to 20 kHz is fine, I just don't think I should need to spend 320 kbits to get it, especially since 160 kbits was giving it to me a few years back. Even stranger is that using 256 kbits with a worse quality factor (q9 for example) preserves more of the upper-end.

I guess what I am asking is how can I get Lame to encode all of that upper-end? Is it not possible in the new versions of Lame? What changed from older versions to now? Is there a command-line switch that I'm not familiar with? Or can I tweak the code and recompile to have it capture what I want? I'm really hoping we don't get into a philosophical debate about spectrum vs. ears, I really just want to know how this can be accomplished.

Thanks for your help. David.

[Slipstreem]

On the basis that what you're asking is, "How can I abuse the encoder and make it do something it isn't intended to do?", all your likely to get is posts suggesting that you carry out ABX tests to confirm whether or not that part of the spectrum is actually relevant to you. Sorry if this seems unhelpful, but it's probably the single most helpful piece of advice you're likely to get as it makes logical sense to determine whether a perceived problem actually is a problem before attempting to fix it.

Cheers, Slipstreem.

[davecool]

How can I abuse the encoder

I think the word "abuse" is pretty strong in this sense. Encoders shouldn't be defended by using this type of language. They are just software.

I think what I was also trying to get at is: Isn't it weird that with the new Lame, even at 320 kbit we don't get the same frequency coverage that we used to get with older versions? Also, the iTunes MP3 and AAC encoders go all the way up to the top of the spectrum when encoding at 256 kbit. Comparing two encoders on that criteria is probably not useful, but at that bitrate, every encoder is apt to be transparent -- so why not capture those frequencies?

I just found another post in this forum talking about the same thing... there are some _great_ screen captures in this post. Shows exactly what I'm talking about. Older lame captured more of the spectrum, new one captures less, no one can explain why.

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. For example, the time spent doing something fun might seem shorter than the time spent doing something boring, but we learned to trust the clock to know both times were the same. I think this is the same kind of thing.

This post has been edited by davecool: Feb 22 2009, 17:29

[Slipstreem]

I think the word "abuse" is pretty strong in this sense. Encoders shouldn't be defended by using this type of language. They are software. I am not the only one noticing this issue.

Call it what you like, but a logical approach would be to assume that the developers have everything right with no additional switches unless someone produces blind listening test evidence to the contrary. I've yet to see any, therefore, it's a non-issue.

I think what I was also trying to get at is: Isn't it weird that with the new Lame, even at 320 kbit we don't get the same frequency coverage that we used to get with older versions? Also, the iTunes MP3 and AAC encoders go all the way up to the top of the spectrum when encoding at 256 kbit. Comparing two encoders on that criteria is probably not useful, but at that bitrate, every encoder is apt to be transparent -- so why not capture those frequencies?

Have you seen how badly most other MP3 encoders compare to LAME at the same average bitrate in the recent ~128Kbps VBR blind listening tests? This may be at least partly as a result of the encoders wasting bits on attempting to encode HF content that isn't necessary.

I just found another post in this forum talking about the same thing... there are some _great_ screen captures in this post. Shows exactly what I'm talking about. Older lame captured more of the spectrum, new one captures less, no one can explain why.

Who cares if the difference isn't audible.

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. For example, the time spent doing something fun might seem shorter than the time spent doing something boring, but we learned to trust the clock to know both times were the same. I think this is the same kind of thing.

So you don't trust your ears to tell you what you're hearing? You'd rather trust your eyes to tell you what you're hearing by looking at graphs? That doesn't sound like a very scientific approach to me.

Come here with a genuine problem and you'll get all the help you need. Come here with imaginary problems and you'll have to imagine answers.

Cheers, Slipstreem.

[esa372]

...we can't trust our bodies...

What a truly amazing statement.

[davecool]

I certainly am surprised to see that the same kind of religious zeal still abounds in the MP3 encoder world ten years later. When I was writing MP3 encoder reviews on MPEG Audio News and Mac Digital Audio ten years back, this was par for the course, but I am surprised that it's still around.

Anyone willing to take a serious stab at my original post?

[Slipstreem]

Anyone willing to take a serious stab at my original post?

My replies were serious! They don't become invalid or worthless just because I presented the truth and didn't tell you what you wanted to hear. This is why so few of us bother to reply to first-time posters asking damn silly questions nowadays!

Cheers, Slipstreem.

[Nick Less]

Have you seen how badly most other MP3 encoders compare to LAME at the same average bitrate in the recent ~128Kbps VBR blind listening tests? This may be at least partly as a result of the encoders wasting bits on attempting to encode HF content that isn't necessary.

Can you please give some examples for these encoders?

Conclusion from the latest listening test:

But again, statistically, they are all tied so there is no quality winner.

[sld]

I certainly am surprised to see that the same kind of religious zeal still abounds in the MP3 encoder world ten years later. When I was writing MP3 encoder reviews on MPEG Audio News and Mac Digital Audio ten years back, this was par for the course, but I am surprised that it's still around.

Anyone willing to take a serious stab at my original post?

Actually, I'm surprised that while the LAME mp3 encoder has improved over 10 years, some people... hmm. Speaking of religious zeal, you display the type which refuses to engage in true, operational (testable and repeatable) science.

Over here you can find LAME versions released in 1999. All the best.

[lvqcl]

I'm really hoping we don't get into a philosophical debate about spectrum vs. ears, I really just want to know how this can be accomplished.

You can switch to LAME 3.90.3 or 3.93.1.

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. For example, the time spent doing something fun might seem shorter than the time spent doing something boring, but we learned to trust the clock to know both times were the same. I think this is the same kind of thing.

In fact, no. You can't tell which codec is better by looking at spectrograms.

[uart]

Hi Dave. It does seem like you're a little more interested in looking at you your music than listening to it, which I'm sure is the thing people are finding objectionable. It’s a fairly common question too. In all honesty maybe the best thing you could to is just try a different visualization.

Anyway the idea behind psychoacoustics is that some information will be inaudible and that is best place to save bits (remember that even at 256kps you're still reducing from 1411kbs to 256kbs). The fact that our sensitivity to sound (as in Fletcher–Munson curves) decreases by as much as 20dB or more in the 16kHz to 20kHz region makes it pretty obvious that sound in this region is the most likely to be inaudible or easily masked.

I think what is happening in the more recent versions of lame is that, even if you manually over-ride the default LPF, the psychoacoustic model will still choose to not encode the 16kHz+ band in some instances. I'm guessing that this is what you were referring to with the word "reliably" in the quote "I can't get it to encode anything above 16 kHz reliably". As in some blocks include frequencies over 16kHz and some don't, yes to the best of my knowledge it's the psychoacoustic model doing that. You know I've looked at the output of other encoders like ogg and aac and guess what, they do exactly the same thing here. If you look at the spectrum over a short time scale then in some places it goes all the way up to the LPF cutoff and at others it stops around 16kHz. So given that these newer and supposedly more advanced encoders do exactly the same thing then maybe it's actually a good idea (though I admit lame seems a little more agressive at it, but probably just because it needs a little more bitrate to achive the same quality level as some of the newer contenders).

BTW. I just tested the command-line "lame --abr 256 -lowpass 19.5" with the latest version (lame 9.8 final-2) and playing back while watching the spectrum analyser I found it was pretty consistently reproducing frequencies up to 19.2kHz (which is the start of the transition band for that filter). This seems to be doing what it's supposed to do to me. Maybe you could test this command line on your material and see how it goes. BTW. The lower the bit rate the more aggressive the psy-model is likely to be here, so abr160 might be "worse".

This post has been edited by uart: Feb 22 2009, 19:25

[Slipstreem]

Can you please give some examples for these encoders?

Looking at the results individually, LAME3.98.2 outperforms the iTunes MP3 encoder more often than the other way around, and sometimes by a considerable margin. The results are similar but not so exaggerated for LAME3.98.2 versus Fraunhofer. It all depends upon how you interpret what you see in the graphs, unlike lossy encoding in terms of listening where what you see in graphs means basically sod all.

Anyway, I've had enough of replying in these dumbass threads started by people who have no genuine interest in the subject at hand. Bye.

Cheers, Slipstreem.

[ameyer17]

Have you seen how badly most other MP3 encoders compare to LAME at the same average bitrate in the recent ~128Kbps VBR blind listening tests? This may be at least partly as a result of the encoders wasting bits on attempting to encode HF content that isn't necessary.

Actually, I thought that everything tied except for l3enc.


(And you could argue from the second chart that Helix did the best job because it didn't have any samples below 4.0.
And the reason why LAME tends not to encode stuff above 16 kHz is that stuff above 16 kHz is really expensive to encode for very little benefit.

This post has been edited by greynol: Apr 2 2009, 21:20

[uart]

Those comparisons don't compare against any older encoders (as in ones that try to always do the full spectrum), so I don't think they really tell us anything about the question at hand. Also (and unfortanately) the low anchor was so horribly broken that it too told us really nothing about aout how (or if) mp3 encoders habe improved over time.

This post has been edited by uart: Feb 22 2009, 20:15

[uart]

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. For example, the time spent doing something fun might seem shorter than the time spent doing something boring, but we learned to trust the clock to know both times were the same. I think this is the same kind of thing.

Ok lets consider that example for a moment. What you're actually implying there is that perceived enjoyment is not necessarily a good indicator of how enjoyable something is and that we should use perceived time instead to gauge this. That really doesn't make much sense.

Here’s a hypothetical study comparing the perceived enjoyment and the perceived time of two different 15 minute activities.

CODE

Activity (15 minutes)          Perceived Time      Enjoyment Rating
Sex with a super-model                12 minutes        10/10  
Slamming your dick in a doorway       10 minutes        0/10

Now the general consensus here is that the "enjoyment" rating is the most reliable indicator of the enjoyment that one is likely to derive from the activity (just like listening to music is the best indicator of how good it sounds). However you are telling us that "we can't trust our bodies" and that the perceived time is a more reliable indicator. That is, given the above data and the chance to choose one of the two activities listed above you'd choose the doorway. Can you understand why we think you're wrong?

This post has been edited by uart: Feb 22 2009, 20:12

[probedb]

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. For example, the time spent doing something fun might seem shorter than the time spent doing something boring, but we learned to trust the clock to know both times were the same. I think this is the same kind of thing.

But if you can't actually hear the differences you can see then I don't get your point. We're not trying to be super human as it serves no purpose with music. What use is it encoding frequencies you yourself have said you can't hear?

If you worked as a CSI or something you might want frequency analysis to find various noises but not when you're *enjoying* music which surely is it's purpose?

[Canar]

I guess what I am asking is how can I get Lame to encode all of that upper-end?

You don't.

Is it not possible in the new versions of Lame?

Precisely.

What changed from older versions to now?

Quality improved.

Is there a command-line switch that I'm not familiar with?

Not AFAIK.

Or can I tweak the code and recompile to have it capture what I want?

Likely.

I'm really hoping we don't get into a philosophical debate about spectrum vs. ears, I really just want to know how this can be accomplished.

Sorry, what you're asking is, frankly, harmful to the audio quality of your MP3s. As we are very concerned about that particular attribute of lossy audio here, the responses you've gotten are to be expected. I am disappointed that people did not directly answer your questions, but not particularly surprised either.

This post has been edited by Canar: Feb 22 2009, 20:16

Reason for edit: Fixed dumb quotes.

[greynol]

Ah, let's tell the guy to use Blade and be done with it; this way he'll get the high frequency response that he craves. Who cares what it sounds like.

[kornchild2002]

I certainly am surprised to see that the same kind of religious zeal still abounds in the MP3 encoder world ten years later. When I was writing MP3 encoder reviews on MPEG Audio News and Mac Digital Audio ten years back, this was par for the course, but I am surprised that it's still around.

Anyone willing to take a serious stab at my original post?

Comparing audio encoding to religion... Now who is taking things too far? davecool is a prime example in that hydrogenaudio doesn't care who you are or what you are, the terms of service still apply. Maybe his highness doesn't like the rules and that is fine, they are free to go post on other forums that believe FLAC hurts audio quality, people refuse to take blind ABX tests because they know they will fail them at low bitrates thus disproving their "credibility," and people are willing to spend $50 on a 1ft long set of stereo RCA cables.

All I can say is that I am glad I never read (or even heard of) articles from MPEG Audio News and Mac Digital Audio. It also seems like there has been an abundance of posts coming from people who just don't want to listen. People wanting to encode at something that is really higher than they actually need (and aren't willing to conduct blind ABX tests), people who bash the audio quality of portable players without any proof other than their biases against the brands, and now someone who thinks they know more than the Lame developers.

Maybe hydrogenaudio should require that people conduct a survey before joining. I enjoy it when people (without certain knowledge) come on the forums and their perspectives are totally changed as they are actually willing to listen. I don't like it when naive people come of the forums who don't listen, aren't willing to change, and think they know more than everyone else (including developers and people who run hardware frequent hardware tests).

[Axon]

Or switch to ALAC, problem solved.

HA effected a paradigm shift on this very topic eight years ago; apparently the OP didn't get the memo. He seems otherwise clued so I wouldn't beat up on him tooo hard.

Also: I was curious to see what Dave's old haunts were all about; looking it up in archive.org, I found paean to the wisdom of r3mix.net and a period encoder comparison: Shows how far we've come in nine years. Or, well, how far we came in five years, because guruboolez pretty much set the standard for how one-man testing operations should be done in 2005.

See also an incredibly misinformed debate about spectrograms on Head-Fi several years ago, and our commentary.

This post has been edited by Axon: Feb 22 2009, 20:39

[davecool]

@Canar
Thank you so much for your reply. You answered the exact questions I asked. So, it can't be done in the current version of Lame. That's fine. I will either need to downgrade my Lame or start using iTunes' MP3 encoder. At 256 kbit I am not expecting any difference is sound quality.

I seriously think, however, that we need to find out _why_ this has happened with Lame. How was this programming decision made? How can it be that even at the highest bitrate available in Lame, it is unable to encode the full spectrum... I mean, even as an option.

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

[greynol]

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

Nonsense!

How can it be that even at the highest bitrate available in Lame, it is unable to encode the full spectrum... I mean, even as an option.

If you can demonstrate that you can tell the difference in a blind test, maybe they'll consider it; otherwise, you're just blowing hot air.

[Axon]

I certainly am surprised to see that the same kind of religious zeal still abounds in the MP3 encoder world ten years later. When I was writing MP3 encoder reviews on MPEG Audio News and Mac Digital Audio ten years back, this was par for the course, but I am surprised that it's still around.

Anyone willing to take a serious stab at my original post?

Comparing audio encoding to religion... Now who is taking things too far? davecool is a prime example in that hydrogenaudio doesn't care who you are or what you are, the terms of service still apply. Maybe his highness doesn't like the rules and that is fine, they are free to go post on other forums that believe FLAC hurts audio quality, people refuse to take blind ABX tests because they know they will fail them at low bitrates thus disproving their "credibility," and people are willing to spend $50 on a 1ft long set of stereo RCA cables.

I do kinda want to know more of exactly what he's talking about when he's referring to "religious zeal" from 10 years ago. Either that means there were a lot more stupid haters/partisans in general, which is true, or people were calling him out for inaccuracies even back then.

[carpman]

@ davecool

What I did to put my mind to rest, was get a CD with content up to 22 kHz. Chop out all data below 16 kHz and give it a listen. This confirmed what I didn't want to believe from ABX tests. There's (generally nothing of audible significance in that range) and it has so little energy that it will be drowned out by content < 16 kHz.

To me the only interesting element to this thread is the psychological effect of looking at the audio player's spectrum analyser and having that "sinking feeling" that something's missing. I admit I've had the same "irrational" reaction. "Oh, it's not all there". And I've had that feeling despite the fact that I know why it's missing, I can't hear / ABX a difference, and I agree with everything the HA regulars have said in this thread. The interplay between eyes and ears and its effect on the psyche is peculiar.

That said, we're talking about audio, so ears trump eyes. For davecool it's simply a case of whether he wants rationality to trump irrationality.

C.

This post has been edited by carpman: Feb 22 2009, 21:10

[Canar]

I seriously think, however, that we need to find out _why_ this has happened with Lame. How was this programming decision made? How can it be that even at the highest bitrate available in Lame, it is unable to encode the full spectrum... I mean, even as an option.

This has occurred through carefully-controlled scientific study, coordinated in no small part through these very forums. That's why everyone's flaming you so hard: frankly, you don't seem to really have any idea what you're talking about.