The -V0 setting increased dramatically in size from LAME 3.97 to 3.98. This seems to be due to "sb21 bloat" or something like that. Can someone describe what this is for the average listener? I finally made the switch from -V0 to -V2 as my file sizes jumped from around 250 kbps with 3.97 to around 280 kbps with 3.98. -V2 seems to be similar between 3.97 and 3.09 ~ 210 kbps. However, does 3.98 offer a significant listening improvement at the -V0 setting vs. 3.97 for these much larger files?

I'm guessing "no", but am really curious what this sb21 bloat is.

http://www.mp3-tech.org/content/?Mp3%20Limitations

Scroll down to Scalefactor band 21 problem.

You may want to consider using the -Y switch which significantly reduces average bitrate at -V0.
AFAIK nobody ever complained about the HF behavior of -V3 which is what you get when using the -Y switch with -V0 to -V2.

The "SFB21 bloat" is not what has caused this increase in bitrate.
There have been both, changes that have affected all the VBR ranges, and tweaks to increase the quality of -V0 (to fill partially the gap that existed between highest VBR and highest CBR).

About the sfb21 bloat effect, the others have pointed you to relevant information about the effect, and halb27 added the notion that using -Y helps preventing that effect to a certain degree.

What I forgot to mention: the higher the resulting bitrate with 3.98 the more is there a chance that the lossless mp3packer process applied to the mp3 file reduces file size in a remarkable way.

Experience with my typical sample set of pop music (guess my bitrate isn't so high because it consists of a mix of newer and older music and doesn't contain metal nor hardrock):

Lame 3.98.2 -V0: average bitrate: 239 kbps
Lame 3.98.2 -V0 + mp3packer: 233 kbps (bit rate reduction is expected to be higher with higher bitrate results)
Lame 3.98.2 -V0 -Y + mp3packer: 217 kbps

It would be nice to know if those repackaged files would still play on Fhg based players (aka "The Nero problem").

I'll try tonight with WMP.

Does your question mean that inefficient packaging of Lame very high bitrate settings is due to FhG decoder's buffer restriction for 320 kbps frames?

Apart from the bitrate increase, is this audible at all ?

What, -V0 vs. -V0 -Y?

ABX and find out. You know that he can't answer for other people.

I just listened to Juli's 'Geile Zeit' with WMP11 without any problem (mp3repacked 3.98.2 -V0 result which has a lot of 320 kbps frames).
I have no experience with WMP though, but what I remember from other threads is that WMP uses the FhG decoder.

I was asking if this issue is audible at all, in common sense (that is, if some folks find it annoying, not me). Nevermind, I always feel discouraged now when I find you already blew in the thread. (You always go over the border with your bluntness). No offence.

What do you expect though? Why would you care if I/we can hear audible flaws in -V 2 and not -V 0 when you will be listening to your music with your ears and your equipment? So what if someone thinks that -V 0 is better with Lame 3.98.2 over 3.97 at various settings? That shouldn't change your judgment of audio quality at all. Hydrogenaudio is a place where people can come to determine things for themselves. It might be nice to know opinions about encoders and whatnot but specific questions like this can really only be answered by the person posting, not us.

So to give you an answer in case your question was about -Y:
I guess there will be persons (best candidates are very young people) who are able to ABX -V0 against -V0 -Y on certain tracks because of different HF behavior.
However I don't remember any thread here concerning this.
Looks like using -Y is a no-issue to at least the the vast majority of people.

well, I think the opposite is the case, or should be, at least.
I think, people apply -V or preset switches simply as they are, with the recommendation not to tweak.
So, vast majority of people is simply satisfied with the lame presets as they are, without additional -Y.

If somebody cuts off the high frequencies , when he uses a high bitrate-quality preset, it does not make much sense (to me).
Such discussion about reduction of few bits is pointless imo (think of todays storage costs, usb, sd-card, HD)
, because , if somebody wants to reduce bitrate, he can simply apply a lower quality(bitrate) preset, be it cbr, abr or vbr, but not tuning with special parameters. If problems happen then with the music, then feedback is wanted here, to improve the settings, encoders.


Maybe the sentence: "Looks like using -Y is a no-issue to at least the the vast majority of people."
is correct if written like:
Looks like using -Y is a no-issue to at least the vast majority of people, who apply it, which is a minority, because most people apply the presets as they are constructed by the developers and recommended.

No doubt, that's the correct version (though the second half is a bit OT).
EDITED:
It should be noted however that a respectable part of the Lame users do have experience with the HF behavior of the -Y switch: at least those who use -V3 and below on a regular basis. Issues in this field are expected to have been reported.
This thread is about sfb21 bloat and using -Y is a possibility to deal with it.
The question whether or not to use absolutely no other option except for -V obviously is relevant to you (and probably to other people), but this question is another topic.

Sorry, but i can't let this go uncommented:

A) the -Y parameter is not a lowpass filter. Nor does it remove high frequencies per-se. The parameter tells the encoder to use a more coarse representation for the higher frequencies, in the parts where it would cause an over-encoding of all the other bands.


B) Also, 50kbps of 250kbps is 1/5th. Not "a few bits".

The sfb21 bloat problem is one of my main reasons to ditch mp3 over aac, since i listen to alot of music that greatly gets bloated because of the sfb21 bug; but for some reason the latest version of Nero AAC seems to bloat like LAME with Metal music. I recken the bitrate increases from 3.97 to 3.98, was due to LAME 3.98 being forced not use the bit resevoir on 320 frames to make it ISO compliant and also tuning on the new psy model aswell.

Anyway i have done a table of bitrates of tracks that suffer from sfb21 bloating; which shows a noticable 60kbps increase.

Thanks a lot for your table.
Bitrate reduction is impressive here when using -Y.

Wow! Wish that was more specified on the Wiki than it currently is.
I did a quick test on John Murphy's "Go go go!" from 28 Weeks Later (very quiet parts and then suddenly goes into distorted guitars and noise)
With -V 0 the bitrate shows to be at 320 for long periods of time, with -V 0 -Y only a few parts hit 320 briefly.
The ABX test basically said I was guessing (by 80+%), so at least I can't hear the difference.

But I love numbers so i did a quick peek in Adobe Audition and noticed something interesting.
(PS! for proper comparison I converted the original to 32bit so the analysis numbers would match the 32bit decoded mp3s)



Not sure what to say as I know a lot of folks on this forum are rater vigilant, so I'll try to avoid claims of quality (as far as lossly goes, quality is perceptive, and with lossless quality is technical).
So facts then...

It seems that -Y might raise the RMS and peaks, thus potentially causing more clipping than when not using -Y.
Obviously a series of tests with varying music is needed to properly confirm this, so... any of the veterans here feel like doing that as I'm a noob when it comes to properly documented tests.

I'm also a bit surprised, as by the sound of it, -Y should make it easier for LAME to properly encode the main data (due to more bits available in the pool), so why is there more clipping? Maybe LAME could use a few tweaks with the -Y code to bring the sound down a bit again?

By the looks of it, not using -Y with -V 0 seem to bring the dynamic range closer to the original, and introduces less clipping.

Personally I don't mind wasting some size if it means less clipping, and like others in this thread, I too am unsure if -Y actually is a benefit or not, my own ABX test showed that I could not hear a difference, but the peaks is a technical concern to me at least. Obviously a good player and ReplayGain avoids any distortions due to the clipping at least.

-V 0 is about max quality possible for VBR right? So if -Y introduces more clipping then it really should not be allowed with that mode, now when it comes to other -V modes I'm ok with that as anything less than -V 0 means you already "allow" a quality drop for gain in size benefits. -V 0 means you tell the encoder to "use as many bits as you need to encoded this at the highest quality you can".

So personally I'm going to avoid using -Y with -V 0 until more tests are done on quality or if the saved bits truly benefit the main audio, or if the "noise" tossed away really should have been there or not. Or if the increased number of clips is reduced in a future version. (at the very least exhibit the same peak behavior as when not using -Y)

Is it possible that whatever -Y tossed away or "simplified" was actually responsible for "softening" the peaks?

I know I know, MP3 isn't exactly the best quality format out there (besides already being lossy),
but it's still fun to try and squeeze water out of a stone right?

With your sample the original is already so close to the clipping edge that it may be pure chance that the -Y result is the one that is clipping more here.

Other than that I think this addresses the general problem that a lossy encoding can clip when the original doesn't. Lowering level by -1.5 db (or more) by using the lossless mp3gain procedure is the way to deal with it (or scaling the input before encoding).

Applying lowpass can easily bring to clipping. Since -Y acts like an "intelligent" lowpass filter, it can also cause more clipping. This clipping is very likely inaudible, but even if it is, it's not a problem for RG-aware players.

In this event clipping should be treated as an artifact. If you don't hear it then why bother with measures to avoid it? It's pure placebo.

My understanding of how -Y works is just about the opposite. With -Y the encoder is allowed to reduce the accuracy of the sfb21 band so that it does not increase the number of bits provided to the other bands, even though they don't need them. Thus there are fewer bits available with -Y, but this should not reduce the quality of any band other than sfb21.