I often encode classical tracks at -V0 --vbr-new. This is a representative result:



It's really rare that lame makes use of 320 kbps. In terms of a quality vbr preset, I think there's some space left, there could be a vbr preset which does make more use of high bitrate frames.

I know, some may say like "-V2 is transparent, what do you want". But I see theres a big bitrate usage gap between the best vbr preset (@--vbr-new) and cbr 320.

There could be a quality vbr presets which results about this:



What do you think?

The perfect solution to that is --abr 260 -b192 or something like that.
This is exactly what I do.
Obviously you don't care about high bitrate but want to get something like the utmost achievable quality while not using cbr320.

With that there's no much use for V0. The VBR modes have the problem that they have to rate encoding quality. This works pretty well most of the time (that's why most people use -V2 or lower). But no software can do that perfectly though Lame is currently improving on it (see Lame 3.98a3). So to me there's no much use in using -V0 as this means very high bitrate while not essantially improving the potential quality rating problem.

abr does not rely that much on quality estimations. It tries to attain the required average bitrate and varies frame bitrate based on quality estimations in a rather defensive way. By using -b n you can further improve the defensive behavior if you're that paranoid like me.
In the end abr x quality scales pretty well with x. The individual encoded track doesn't have a big influence on used bitrate which is not very intelligent on one hand but very safe on the other.
If you're out for high bitrate this is the best way to go IMHO.

I can imagine a superior vx behavior which uses an increasing degree of safety while reducing intelligence with decreasing x such yielding with v0 a defensive behavior pretty much the same as say abr260 but giving a lower average bitrate than 260 kbps when it can do so in a very safe way.
But as we don't have this behavior using high bitrate abr is the best thing to do IMO.

As a sidenote: I prefer 3.90.3 non-alt-preset high bitrate abr as this yields good results with the worst problem samples I know apart from overall great quality.

Hm, this is my opinion:

If -V0 does not allocate 320 kbs frames on your music and you try to force Lame to do so anyway, you are in fact wasting bits

You should be happy that there are no 320 kbs frames necessary in your music. In fact, I'm always a little afraid when Lame allocates lots of 320 kbs frames, because that might mean that some frames even needed more bits. Just look at it that way!

a) what Lame version are you using?
b) what kind of a track is this?
Looks like it's a rather easy one for encoding.
Keep in mind that it's only necessary in very difficult situations to use 320 kbps frames.

a) Lame 3.98a3
b) Classical (Track 2 on Bruckner#9/Wand/BPO/Sony RCA)


But this is by no means reflected in the two bitrate histograms. One does make use of 320 @ 0.5%, the other @ 36.1.

I don't get it, in your first post you have a chart where you have plotted a nice bit distribution graph. I get the feeling it's by the way it looks the best, and thereby sound the best. Right? I mean, aren't you more interested in the way it sounds? Or at least the bit rate? If you want higher bit rate, the last thing I would come to think of is distributing the bits by the way it would look on a chart! (i.e. a nice, even, symmetric stair case )

If you really want more 320k frames, try tweaking with the lowpass setting. Your music probably doesn't have much HF content, so LAME does not need to allocate as much bitrate.

The plain simple point is that if you care about rare messups in the psymodel, then you dont trust the psymodel of the encoder - and therefore should encode everything at 320kbit. Everything below that, will result in some ultra-rare barely-noticable artifacts. Thus, VBR is the wrong tool for the result you want. But wait, even at 320kbit, there are some ultra-rare cases where its not enough..... so you should instead use a lossless-encoder.

Short version of the story: if you want perfection and 100% safety, you should not use a lossy encoder. Bitrate-distributions and all kinds of other graphical representations, will also *look* much better if you go lossless. Of course, all this is completely ignoring the question if you can actually notice the difference at V0 during everyday-listening - but well, the goal of the whole proposal wasn't about real-life performance anyways, but instead about "feeling better".

Lyx - you have a lot of useful information and I'm pretty sure you are quite knowledgable about this stuff. If you scaled back a bit on the snarky meter, that information might get across.

I've been encoding a fair number of LP transfer albums at -V 0 --vbr-new simply to make secondary backups requiring less space than my primary lossless backup. I've noticed a great deal of 320 kbps on quite a few tracks, but none on others. I also notice a fair number of tracks encode with close to 100% LR stereo. My casual observations is that both the high bitrate and lack of MS stereo are different than 'average' tracks from the commercial CDs I've encoded.

Maybe this isn't the best place to insert a question I've been wondering about, but here it is anyway.

A few years ago I encoded several dozen albums with the Frauhofer facilities of CoolEdit 2000. I aimed for the highest quality it could produce so chose the CBR 320 preset, which specifies a bandwidth of 22050Hz.

I know LAME has a lot of optomized coding. Much of that is in the 'standard' preset to achieve both high quality and less storage requirement. Has any of that effort been extended to CBR 320?

Obviously the space required could not be reduced but has there been development there on improving the audio quality? Would anything be different for those albums I encoded at CBR 320 if I had know about LAME, or if I did them now with LAME, regardless of whether or not I might hear that difference (not that I actually contemplate redoing them, just curious).

Regarding the original question, yes there is a VBR step missing after V0.
I know that. However, we only have a 10points vbr scale, and we made our best into fitting it to to most usefull steps, so this extra high vbr had to be left out.


castanets, fraftwerk

You could incorporate a high quality preset above V0, -VX, and do some marketing hype (I know, lot's of us are not the friends of marketing) around it

I am wondering if the LAME development team are considering adding an extra setting between -V4 and -V5? I feel that the jump from ~130 VBR to ~165 VBR targets is quite big. Especially considering the results of the recent ~128 Kbps VBR test. By that I mean that -V5 is nearly transparent, but not quite. But it seems that -V4 may be too big a jump than necessary, for example for portable applications. It may be useful having a setting that targets around 140 - 150 VBR. especially for players that don't support AAC or Vorbis.

Sorry about this being somewhat off topic.

Regarding the original post. I think -V2 --vbr-new sounds great on everything I've tried it on. When I encode a song and it doesn't need 320 Kbps frames I'm happy!

That's exactly what V4 does on my music (3.97b2 at least). The songs I encoded where all somewehere in the 130-160 kbs range.


That's exactly how I think about it

I'm recording using -v3 Vbr New. I did some ABX with my symphonic music and couldn't tell a difference between the original CD and the mp3 copy. My ears may not be good enough to notice the artificts that people talk about. I have a 60GB Toshiba MP3 player and I can put over 830 albums on it. I'm not sure why people need 320 when my copies at approximatley 155 bytes are clear as a bell. Anyway, I like -v3 vbr new. I suggest people try it and see if it's all they need.

Please don't get worried by people like me who may be a bit too much on the perfectionist's side. But I can't go against my nature, just try to behave reasonably within the framework I'm in. Which is a matter of taste of course.

[quote=kenny01,Feb 11 2006, 11:16 AM]
I'm recording using -v3 Vbr New. I did some ABX with my symphonic music and couldn't tell a difference between the original CD and the mp3 copy. My ears may not be good enough to notice the artificts that people talk about. I have a 60GB Toshiba MP3 player and I can put over 830 albums on it. I'm not sure why people need 320 when my copies at approximatley 155 bytes are clear as a bell. Anyway, I like -v3 vbr new. I suggest people try it and see if it's all they need.

[/I understand the quest for perfection. It's just that I'm trying to get the best qualtiy for my MP3 player and put as my of my albums on as I can. The quality of MP3 has improved over the years. Even at 128 I can't really tell a difference between the origional and the CD. I'm sure there is a difference, but it's so slight it really doesn't affect enjoyment of the music. It is the 'quest for perfection' that seems to make this hobby so interesting and also the motivation to make the MP3 codec even better than it is.

I guess that Ogg Vorbis has a huge advantage here in that you can add the decimal place into the equation, and thus assign settings such as -q 4.5, 4.6 etc etc, that give smaller incremental steps up/down in the quality range.
I assume that this kind of lower-level, more "detailed" setting is not something LAME (or mp3?) could do? or is it that to do that would be a total rewrite of the LAME encoder?

Please continue the problem samples discussion here:

ABR vs VBR, Problem LAME Samples.

Apologies if the topic split is not seemless, I'm still learning.