So, why is .mp3 capped @ 320 kbps?

Would be nice to have an option for say 400-500kbps. Would be a nice size/quality ratio I think.

I don`t know everything about .mp3 so no trolling about what I`ve should have know or similar. Enlighten me instead;)

thanks

Could you please share some samples with us where 320 kbps mp3 produces poor quality according to your hearing ?
Or to refine the question, for what purpose are you planning to use mp3 at those bitrates, instead of lossless compression ?

From Wikipedia :
http://en.wikipedia.org/wiki/MP3

http://mp3decoders.mp3-tech.org/freeformat.html

With Lame and the --freeformat parameter you can encode up to 640 kb/s, but the compatibility with most decoders is broken.

EDIT : beaten by fred_frno

Nice size/quality ratio? I'm not sure about that. You can force LAME to encode at higher bitrates using the freeformat switch. But you shouldn't be doing this because of

* Nearly zero support for freeformat streams

* But more importantly: It's not clear that MP3 does scale that well for upper bitrates. The fixed filterbank and fixed codebooks have been optimized for lower bitrates. So, you might not experience the expected quality gain for a given increase of bitrate.

If you want to have 500 kbps streams you might be better off using WavPack lossy or something like that.


Edit: Darn .... Am I that slow? I hit "reply" when no reply was visible.

I agree. With some problematic samples (castanet), --freeformat -b 640 is not better than --preset insane because pre-echo problems is inherent to MP3.

The optimal size-to-quality ratio is actually the lowest bitrate at which *you* can't tell the difference between the lossy file and the original source (i.e. the mp3 is transparent to you). Example: For virtually all the music I listen to, I can't tell the difference between the orginal CD and an mp3 encoded with LAME at -V 4 (VBR, varies around 165 kbps). For me, any higher settings/bitrates would be wasting bits and drivespace on audio I can't hear.

Thanks for the replies and for keeping it civilized and educational!

I can tell (or think I can) the difference between vbr encoded mp3 and lossless on my ipod. Not for all kinds of albums or styles but for some.
And that`s really all that matters.

For instance, for those familiare with it, I think the latest Tool alb. sounds better(more punch, crispier cymbals etc) in lossless than mp3 on my ipod.

The problem is obviously that lossless sucks a 10 gig ipod free of space fairly quick, hence my question.
I was trying to find some golden, middle way;) Is there any?

I won`t dabble with the freeformat stuff.

And no, I don`t think a 320 or vbr ex. sound bad...just think it can sound better in some cases. It`s still compressed.

Im suprised no one has posted yet suggesting you try to blindly test if you can really hear the difference, especially on an ipod which doesnt come with the highest fidelity headphone. Try an ABX test (search the board - its everywhere here). Personally I use ~115 vbr LAME on my ipod and probably could drop to ~100 and be just as happy. I keep everything backup with FLAC and transcode even though I cant usually hear any difference.

If you want absolute quality, I think you should just go lossless. At least that way you can have no doubt in your mind that the fidelity hasnt been touched with an sonic-altering algorithm.

Just enjoy the music bro,
-Joe

IMHO, your iPod just uses different postprocessing for mp3s and for lossless files. There's a possibility that with lossless you have some replaygain, dithering, noise shaping, etc. On top of that, mp3 decoder may not be of highest quality... Can anyone, who knows more about iPods confirm or deny my thoughts?

The first thing I did when I bought the ipod was to toss out the horrible earbuds that came with it. (For those wondering Im using porta pro & a closed phillips headset. The porta pro sounds pretty good now after years of use hehe. The phillips set is quite new so it still a bit "hard". Been pondering on something better but im not made of money)

For mp3s i use lame vbr new V0,3.97b2, and lossless is made from flacs to wave to wma lossless to apple lossless (a lenghty prosess but it works. Def. open for suggestions on this one; how to get flacs to the ipod?). Not sure if it involves postprocessing of any kind.

Haven`t done the pepsi challenge. Don`t think i will.

One other thing is that i think a lot of jazzalb. sound better lossless than compressed. This is in a noisy environment such as the subway etc and i need to turn the volume up.
Not saying this applies to every alb known to man though.

As Erukian said, I might just go lossless all the way.

And most definetly enjoy the music;)

You initially said nothing about perceived quality (as in "this sounds better than that"). But since you just did let me say I agree with Eli's suggestion: Try doing some blind tests to prove to yourself that you hear differences. This is a great opportunity to check what quality level you really need for your iPod. I'm quite happy using -V4 for my portable.

Then don't expect anyone here to believe (or care) when you say something "sounds better" than something else. I think you may have fallen into the trap of the placebo effect - lossless is exactly that, lossless, and most of what you think sounds "compressed" is likely your personal bias, since you're already aware that you're listening to something that's been compressed.

Have you read the terms of service, specifically TOS #8?

http://www.hydrogenaudio.org/forums/index.php?showtopic=3974

Mmm.... Could there be a chance that what you're hearing is clipping? being an -V 0 --vbr-new encoding, it is quite surprising to find a difference.
Yet, "more punch, crispier cymbals etc" is not a description of any artifact that MP3 adds. At much, there's the possibility that it's pre-echo, but that should be verified. Else, we have to assume that this is a moot case.



This especially puts us in the perspective that you simply don't care, and imagine things.

As SebastianG said if you're out for best quality and allow for 400-500 kbps you're probably best off using wavPack lossy which you can use with your iPod if you switch to free Rockbox firmware.
With this you can be pretty sure there will be no audible difference from the original.
And you can allow for lower bitrate: wavPack @ ~ 350kbps is sufficient IMO even for extreme quality demands.

I did similar considerations concerning mp3 once and wanted to use 320 kbps. But I realized using ~250 kbps yields practically the same quality. Perfectness is not achievable with mp3 but a a very very good quality is. As for VBR I share your considerations. In the very high bitrate range it doesn't offer advantages but provides for a certain danger that it makes things worse (though this is rare). To me Lame's ABR is the best way to go in the very high bitrate range (though there's nothing wrong using cbr 256). As for the encoder in the very high bitrate range I prefer old Lame 3.90.3 up to now because of it's quality robustness.
I'm with the other contributors who say transparent quality is usually achieved with something like -V4. Unfortunately this is not valid for the universe of music. harpsichord for example requires a higher bitrate for very good quality (abr 224 or better IMO). But with a good encoder in the 250 kbps range you're pretty safe (with the exception of perhaps pre-echo issues in case you're sensitive towards that which you can check with samples like castanets).

Does it come with s/w? (link?)

I`ve settled on vbr V0 for rock & pop, i realized i can`t justify the filesize/quality ratio. Some albs i still think sound better with lossless but it`s to few to have them all be lossless.

If the firmware will let me use the pod the way i want to i might consider switching.

With jazz I`ll go with lossless. The mp3s just make the horns sound too hard on loud volume.

Again, thanks for the replies!

"Breaking in" headphones is pretty worthless. From how I understand it, the diaphragm and the surrounding materials may be stiff and may ease up over time, very slightly changing the sound from a new pair and an old pair of the same cans. Simply playing some bass through them for a few hours is sufficient. A multi-year breakin regimen is nonsense.

What?! Why not just go from CD (or FLAC) to Apple Lossless directly. All that extra steps and conversion can only introduce extra room for error.

That statement means nothing to anyone on this forum and just makes you look like you like to say things without proof, unless you can provide an ABX test from the original vs. ~320kbps MP3 that confirms what you think you hear.

Are you serious? Not everything is science! All that really matters is what you hear isn`t it!?

Lot of stuff is not directly from a cd. With cds I own I use that method yes.

Yes, and the problem is that most people have no idea what they actually hear. Thats why ABX is required around here when making statements that are contentious. Fact is most people have no idea what they can hear, in spite of what they may think.

Hearing is a matter of perception; but when it comes to compare two different signals you have to listen them separately. That's why any audio comparison involve not only your perception but also your memory: comparison is a confrontation between a present experience and a past one, between what you're currently hearing [experience] with what you heard previously [reminiscence]. But memory is faillible. That's why even trained and professional people are able to detect difference between A and... A. And that's why ABX is often required to validate any comparison.

Can I ask a question.

Is it possable to do VBR above 320kbps? Sorry if its already been asked, I must have missed it!

Not with MP3. --freeformat (bitrate up to 640 kbps) is CBR only (at least with all existing encoders, but I'm quite confident that MP3 specifications don't allow VBR + freeformat together).

Check out Rockbox and WavPack... and Google for future reference.

Interesting, you get flac support (and more) with this firmware. Too bad I`ve got the 2nd gen ipod.

If you can't use Rockbox firmware and demand for best quality for your iPod IMO you're best off using very high bitrate aac. Second best solution IMO is using Lame 3.90.3 abr ~250 kbps or cbr 256 or api if you are extremely cautious (using lame 3.90.3 is subjective and there are good alternatives for sure).

But as so many contributors said you should be very suspicious towards what you beleive or think about your abilities to hear differences. ABXing isn't perfect but it's the only solid way to talk about differences. And it's easy using foobar (but it's pain cause you will provably find out you don't hear any difference between say lame abr ~250 kbps and the original).

And what about 3.89 --r3mix?
May I suggest you to recall each time you're claiming the superiority of your favorite setting that HA.org recommandation is totally different than yours? I don't think you're really helping people by recommending a 5 years encoder, with settings that Dibrom (aka the guy who tuned --presets in 3.90 with many samples) himself refused to consider as optimal, without mentionning that nobody here ever agreed with 1/ your methodology and 2/ your perception [just to quote recent exemples].

Yes, I'm not with the current HA recommandation in my application field of using very high bitrate. And I've never seen members who are engaged with the current HA recommandation being engaged in the very high bitrate field. So I don't feel doing wrong especially as meanwhile I am used to talking about my favourite setting in a way that makes the subjective nature pretty transparent. Usually I am positively talking about the good progress of current 3.98 alpha development which I can clearly see. But I don't see why I should be quiet about my experience. First of all 3.90.3 has been the HA recommandation up to relatively recently. And I also think there's no doubt that with the 3.90 branch (the different 3.90 versions as well as 3.91 which are more or less identical except for special items) there is a wide-spread long-term positive experience which is value in itself. This doesn't mean there's no room for improvement, and I'm sure there was a significant progress in the moderate bitrate range - that's why I'm totally with the current HA recommandation in the moderate bitrate field.

As for practical experience in the very high bitrate range: the usual experience is that there is no difference between the various lame versions - they're all just excellent. But when going for best quality in extreme case there are differences though nobody can talk about an encoder's behavior towards the universe of music. There is the pre-echo issue immanent to mp3, and from the very limited experience published here it looks like 3.90.3 is behaving better than current lame (see problem sample thread). BTW that was the very reason why I switched my favorite setting from gpsycho to nspsytunes. More essential to me however are 'my' three bad tonal samples harp40_1, trumpet and herding_calls, and with this up to now 3.90.3 is way ahead (when avoiding VBR which has it's own issues in the high bitrate range).

Because of the quality achieved at very high bitrate we luckily don't have a lot of practical experience with problems so it's correct to say methodologically speaking my meaning is not relevant. So everybody is quite alright ignoring the limited experience available, but everybody who doesn't isn't wrong either.

BTW I really wonder why we don't have significantly relevant experience with pre-echo samples of different encoders at very high bitrate. As there are sufficiently many members who are sensitive to these problems meaningful results should be attainable.

You haven't searched very well. As I said before, the founder of this board, who also tuned the LAME encoder you're using, put most of his effort to bring to the community high bitrate presets. Old members can still remember the --dm-preset era Dibrom tuned before HA.org was started; then appeared --alt-preset standard (~200 kbps), --alt-preset extreme (~250 kbps) and --alt-preset insane (320 kbps) which were the most strong, reliable and famous tunings brought by Darin (Dibrom) to the MP3 community. And do you know what this developer --who shared the same interest for HQ encodings than you but put much more energy in this passion than you-- said about CBR you've just recommended before?

http://www.hydrogenaudio.org/forums/index....ost&p=18573

I can't find anymore the list of most samples used by Dibrom to tune LAME during the maturation phase of --alt-presets. But it was far wider than the one you're constantly using for your comparisons. So don't be surprise if his past conclusions, confirmed in the past by several members, are completely different from yours. You should therefore question yourself about the cause of such difference and investigate further the validity of your method.


Because your experience is based on a too limited and biased set of samples
Because you have troubles to detect some elementary artefacts like pre-echo
Because the conclusions of your biased experience are going against the conclusion made by the community.

Experimenting on your side rather than following the recommendation is an excellent thing -- the best people can do. But you're not only wasting your time by using a faillible methodology, you're also fooling people by recommanding settings (they are often changing it seems) based on a such wrong method.


It's good for you to recall that 3.90 was recommended during a long time. The weird thing is that during four years nobody ever noticed that --preset standard and --preset extreme were inferior to the --abr encodings you're now promoting for a year (with some variants). It's either that hundreds people were totally deaf or that your conclusions are, well, very personal...


You said it: the available experience at very high bitrate is very small. A bit short for make conclusions, don't you think?
And I also recall that the available experience is sometimes going toward your own findings (like here). Did you forget when you've recommended your ABR/CBR settings last hour?


It's not very hard to explain. There's few collective interest for tuning lossy encoders at very high bitrate. First because it's very hard; second because most people interested by transparent encodings are now hesitating between flac and wavpack rather than between LAME 3.90.3 or 3.98. HDD weren't as large as today when the main HQ lossy encodings projects (--alt-preset, MPC) were active. The remaining people who are still looking for HQ MP3 are also fully happy with latest LAME builds: they are all offering the same perceptual quality than past reference encoders and they are also more pleasant to use (faster encoding speed, more options and efficient settings, etc...).

halb27
Could it be that your music collections consist of LAME 3.97 killer samples?

I've never heard your own sample that benefit from superiority of 3.90.3. If no improvement on your music, why do you care only 3 exceptions.

AFAIK, there is no quality difference between LAME 3.90.3 and 3.97... I may be misinformed, but I haven't heard of samples that sound poorly on latter and great on the former. It might have been the case with 3.96.1 (I remember reading a thread about it).

haregoo
how much LAME 3.97 killer samples are out there? Can they even make a decent music collection?

Anyway, back to the topic, is there actually an objective need for bitrates higher than 320 kbps? Personally, when I use LAME, I use -V 0 and I can't hear the difference between that and 320 kbps. I remember hearing once here on Hydrogenaudio something like this (mea culpa, I didn't remember who said it) - giving mp3 a higher bitrate than 320 kbps is like taking an old Audio 80 and sawing a hole under the gas pedal so you can press it even better... Yes, it will burn more fuel, but it won't go any faster.

Which codec (apart from lossless, of course), would actually be worth giving bitrate above ~300 kbps? Does Ogg Vorbis @ 500 kbps make any sense at all? What do the listening tests say? I know Vorbis is transparent for most samples/users @ ~160 kbps, and that's exactly the reason I ask this.

edit: typos and a question

Two specific problems were discovered with LAME 3.97: noise with tonal samples; noise due to an insane usage of short-block with unexpected samples (trumpet.wav).
These problems must be fixed (the first one should be solved with 3.98 alpha). But these samples can't be used as sole source for any listening comparison.

I sent 33 or 34 samples illustrating the first problem to LAME developers (which were used by myself to check how efficient was the fix introduced in each alpha). It wasn't hard: "tonal" instruments (various wind and string instruments) and therefore "tonal" samples are quite common with classical music. Then yes, this problem can affect a large collection


Pre-echo was known to be a real issue with old Vorbis encoders, even at high bitrate. But even with pre-1.0 version I wasn't able to ABX castanets samples at 500 kbps. I don't think that 500 kbps encoding for stereo material was ever necessary. I guess that such bitrate exists for people who don't trust, or like, or know about listening test (i.e. most "audiophile" people on the earth).

...
Please keep to the facts. My personally Lame settings in the very high bitrate range did change but not the way you give the impression of.

I started with cbr 320 but relatively soon found this isn't necessary for me and went over to abr 270. The pretty questionable thing I did (because of ignoring the joint stereo advantages to a great extend) was to use gpsycho abr 270 for a long time. Since quite a while I use --alt-preset 270.
The other things (lowpass and - for a long time - lower limit on frame bitrate) have never been essential to me and I've always made that clear.

As for abr ~270 or cbr 256 I prefer abr but from my personal experience cbr 256 is just as fine. Except for pre-echo samples I am not afraid of any problems with such a bitrate just as most people aren't afraid of using -V2.

As for dibrom's words on VBR it is not clear which bitrate range he had in mind. We really shouldn't generalize 'VBR vs. ABr' or things like that. Usually things depend on the application context. In the moderate bitrate range I can understand him well, at 270 kpbs I don't. Most people think like: -V0 has an interesting average bitrate and provides for best quality that is not superceded by going higher in average bitrate using abr or cbr'. But in order to be true VBR has to be very robust. Which isn't the case up to now though I can see the progress in the 3.98 alpha branch. And yes: to me few samples are sufficient when I can't expect more differentiating samples coming up.
I'd also prefer -V0 or even better -V1/-V2 over abr 270 if I could rely on it. Maybe this is the case with 3.98 final.


With this in mind you should understand my preferences in the very high bitrate range. 3.97 still has the first issue you mention, and the 3.98 alpha branch hasn't reached the end of the road. As for pre-echo samples we unfortunately don't know a lot about different versions' behavior in the very high bitrate range because of the fact that few people are interested in that bitrate as you said. The few things we know however are in favor of 3.90. As for "normal" music I think we both agree that in the very high bitrate range it doesn't matter what lame version or setting to use. At ~ 250kbps they're all excellent.

Your merit is that you are able to contribute to current Lame development in a very constructive way. May be my posts here did contribute too towards Lame development, but they did and do in a way that looks a bit destructive rather than constructive. Maybe this bothers you. Maybe not many people think like me: With my job (developing database applications) I always prefer if my work is so good that no errors come up. But of course errors do show up, and when they do I think like this: I would have preferred if this error would not exist. But obviously it does exist. And if it does the best thing is that it shows up early. This very attitude motivates me to try to produce less errors, but at the same time makes peace with real life where errors exist. And usually it is other people who show up my errors. I like that - the background is always: if an error exists it is good to be shown.