Is there going to be any difference in transparency in any files?

and i've been reading threads all night... someone do me a favor and quickly explain why mpc is better than mp3 (like the specific fault of mp3 that mpc doesn't have?? they seem very similiar)


i'm 99.9% ready to go to mpc as my format of choice... its the best balance of quality/availability/file size... its either mpc or mp3... and since mpc makes smaller files... i really can't think of a reason to stay with mp3 aside from compatability....

The real competitor to lame --preset insane is not mpc --braindead, but mpc --standard. There is no need to increase too much the bitrate with musepack, if you want to reach transparency (or near-transparency).

Mpc has a better psycho-accoustic model. A better time-resolution (no pre-echo on sharp attack : drums, electronic), less distorsions on some special case (harpsichord). Mpc is gapless - mp3 isn't (important for live music, opera, etc...).

Therefore, if compatibility isn't too important, there is no reason to stay with mp3. I switched to mpc one year ago : I never regret it, though I have a mp3 portable player.

yeah, i did a lot of reading about the different bitrates of mpc... i basically want to ensure that everything i encode is transparent (or as close as i'm going to get) without having to change the settings each time...

i figured braindead was a safe way of doing this... and since the filesize is still smaller than aps insane, i chalked it up as a win...

in reality, filesize is not a priority to me to an extent... ie... my dream situation is lossless with 20mb or less per track on average.... but right now thats not attainable (atleast that i know of...) so i just want to get the best quality from a lossy file as possible... a meg or 2 here or there isn't important to me... but at the same time i can't afford to have a full cd be more than like 200mb, or its just not worth it...


but if i'm understanding this right mpc standard = aps insane... mpc braindead is just better... ?

Braindead is not as tuned as Standard so you should stick with Standard.

oh... i did not realize this...

The rare artifacts in a Standard encoding will more than likely appear in a Braindead encoding.

With mpc --standard, you should encounter less problems than with lame@320.

mpc std is not eaqual to lame insane. Two different formats can't bring identic results. On critical stuff, lame (and mp3 in general) would fail. And mpc not. But there are some rare samples (amnesia.wav) where mpc --standard (1.14 codec) is not as good as usual, with some flaws that mp3 doesn't show.
If file size isn't too important, I suggest you to encode with --extreme, or more, if you really want to feel safe with your encoding.

Many of the MPC users of this forum feel that there is no point in anything above Standard.

does extreme have the same problems as braindead? (ie, not being well tuned like standard)



and i'm using the phrase "equals" VERY loosely... i'm talking in general terms of coarse...

The reason Standard is more tuned is because it is the most used setting among the MPC userbase. I've found most MPC users who use Extreme are those that want to transcode their MPCs --> MP3s because Extreme will handle this better than Standard.

I don't know that there is any evidence that extreme, braindead, or any of the higher-quality settings are "less-tuned" than standard. Sure, maybe standard required more tuning to attain its degree of transparency, but I don't think there has ever been a scenario where extreme or higher failed on a sample while standard didn't. In short, all the quality levels including and above -q5 should be safe to use. Whether you need to use higher than -q5 is a different discussion (with many threads on the subject available with the search function).

Sorry. But even with « general speaking », it's difficult (for me) to agree on such equations.

I'm not totally agree with the futility of an --extreme, or --insane encoding. The rare artifacts we can hear with std are progressively reduced by higher settings. There are some little difference I can ABX at --standard, but not at --extreme. On some particular case (harpsichord : I'm crazy about this instrument), some days, I can ABX up to insane (after listening >100 times the same sample). I never succeed with --braindead.

Consequently, there is a real and virtually audible progress.

Interesting points.

[ 1 ] When choosing a quality level, remember that --standard is made to use and exploit the correct, unbiased settings for the trained ear - along with an reasonable margin of safety. This means if the codec was well made (it is), all differences should be above the capabilities of the ear to perceive.

[ 2 ] Using anything over --standard will just thicken the safety margin, thus buying a bit more insurance against small quality discrepencies in rare passages, DSP's or alien-like hearing. Of course, thickening the safety margin will usually cause a significant waste of bits. By the way, if musepack was perfect, everything above --standard would be 100% useless. In practice, since musepack is not perfect, let's say that 95% of it is useless.

[ 3 ] Very important: AT ALL QUALITY LEVELS, THE CODEC ANALYSES THE SOUND IN THE SAME WAY. This means, if it happens to underestimate the necessary SNR, TMN or NMT thresholds at a given time for a given range of samples in a given frequency band, it will basically do so at all quality levels. This is why going over --quality 5 will usually just soften the problem, not solve it. In short, one can say that all settings above --quality5 just add a bias over of the calculated transparency parameters. At the end, if a sample really trips up musepack, there's really no guarantee to stay protected enough at --braindead.

[ 4 ] Unlike mp3, musepack was never made ("marketed") for "good enough" quality at low bitrates. It was designed to break with this trend.

[ 5 ] I don't recall anybody hearing anything annoying on *any* sample at --quality5.

[ 6 ] An interesting example: let's say that mppenc 1.14 miserably fails on sample "A". It's perfectly possible that the future mppenc 1.16 solves the problem so well, that: mppenc 1.16 --standard is better than mppenc 1.14 --braindead ! Bottom line: correct psymodel can be much better than failing (in this case) psymodel with a big margin of safety !

For all these reasons, I recommend for the original poster to consider --standard (= --quality 5) or - why not - --quality 6.

In my opinion, anything above -q6 is either:
- targeted for extreme, thorough tests on special samples for many hours, like guruboolez
- lack of trust in the codec developer.

By the way, choosing any setting above --quality5 will never decrease the quality. As far as I know, mpc is fully threshold- and parameter driven. For this reason, even though the most tuned setting is --standard, all above settings will benefit from improvements on it. Also no compromise will kick in at higher quality levels, and there's no bitrate ceiling (except for worst case, near-lossless ~1400kbps = ~16 bits per subband sample). Please correct me if I'm wrong here.

I use q6/xtreme just for the extra margin since the bitrate doesn't make a substantial jump over -standard. I used q8/Braindead for a while, but after determining that I couldn't hear a difference over q6/xtreme, I decided I'd rather have more room for audio than have the extra bits.

If nothing else "xtreme" just sounds cooler than "standard".

For the sake of truth, there have been recently some discussions about some small problems with --standard with some very special electronic music samples. I don't know, however, it those can be considered "annoying", since I have not listened to them.

With any lossy encoder, there's always a possibility of not being transparent on one particular case.

Do you think it is safe to go below --quality 5 ?

At work I've got a very small hd, and wanted to use --quality 4 or even --quality 3 just to save some space.
thanks

bye

Yes, there's always the possibility of small differences, for sure.

My opinion: in many cases,
- the difference won't be disturbing
- If it's on the edge, you can go for --extreme
- If it's serious, you can't really fix it.

All in all: pretty much all of these electronic samples are death for mp3.

I've tried --quality 4, it still sounds pretty good but I personally wouldn't go below that. On many samples, with average pc speakers it seems 3.5 will be okay.

Edit: spelling

Tried the same and hear no difference between q5 and q4 (put it that way, with the background noise of a common office and my cheap speakers i wouldn't tell the difference even between cd and --quality 2 )
Thanks.


bye

Try under 3.0, then you can hear the "piou-piou" sounds that guruboolez hears up to --quality 8.0

Can't hear it.
Whoa i'm deaf

Strange ! Maybe I'm a bit over-sensitive to this.

I'm lucky it's only up to --q4 for me though. It'd really be a pain to have to use --quality 11 like for guruboolez

Just joking a bit... mpc is trasparent to my ears at q6, and it's really hard to spot at q5.

Anyway i'm interested in the artifact gurubooles hears... i've searched but can't find the post, have you the link? (...only if you got it easy, don't waste time searching for me )

Thanks a lot


bye

I suddenly perceive a warm fuzzy irony in this topic...

Apart the joking, if you refer to me, i'm truly interested.



thanks


bye

You could. mpc in not bad at lower bitrates, but Vorbis would be a better choice here.

With the new PNS being used at q < 5 mpc sounds much better than before in this bitrate range. But in some rare occasions it fails badly and adds rasping noises, so I wouldn't really call it safe to use yet. But for causual listening you may not care about these rare hickups. You just have to try it and see for yourself if it's good enough.

To be honest, I switched from braindead to insane, but lower I won't dare to. This has nothing to do with trust or something like this... I simply love overkills. Just the feeling, I did my best... That's why I encode for my friends mp3s with insane and lowpass 21 (in extreme cases I switch down to extreme with -Z and lowpass 21)... they hate me for it and use four letter words sometimes, but later they come in and are very thankful and talk it up; that's why I prefer HDCDs over CDs

The piou-piou is a reference to the particular artifact of musepack with a sample like castanets.wav. Not really pre-echo, as other lossy encoders, but a kind of inner deformation.

The exact link, to a french forum, is : http://forum.hardware.fr/forum2.php3?post=40367&cat=3

Thanks a lot!

bye

/edit
hope google can translate this...

Back in mpc 1.14, I do be (I do am ?) annoyed by what happens to amnesia.wav at quality5. The problem persists in quality 6, and is gone at quality 7.
Improvement in 1.15r, but I'd have to listen to it again, at home, to see if I still consider it to be annoying.

No kidding..


guruboolez, glad to see you have taken no offense from my childish teasin'

Seriously though, I really appreciate your insight in listening tests. Do you think that "piou-piou" you can sometimes spot at higher bitrates, is within the same family as ice-related sound I get under --q3 ?

It seems to be a well-balanced distortion: a bit of noise, sometimes a slight bit of ringing, almost completely covered by flangering. Overall it happens with most (even slight) transient situations, --quality 3 and less, bitrates usually < 70kbps (IIRC).

Besides - I'm curious to know if you have tried tweaking 1.15r using new short-block commands from info.txt ?


Ok, I stand corrected. It seems like a tougher sample, haven't tried it thoroughly, despite the big discussions on it. Will be trying next weekend, we'll see

Have you tried Frank's new experimental MPC encoder that I posted about in that thread? It's supposed to address these problems directly..

It almost seems like everyone who was talking about the problem lost interest the moment Frank tries to fix it

Yes, let's all try these beautiful new commands: http://static.hydrogenaudio.org/extra/mppenc.info

The suitable mppenc version can be found there: http://www.hydrogenaudio.org/forums/index....12&t=6558&st=50

Edit: Dibrom ? There seems to be a bit of HTML garbage when editing my latest post (before this one), do you know what it might be ?

I'm trying to get at least 4 hours of sleep per night, but I don't always manage it

No... it looks like a bug or something. Maybe a corruption in the database, not sure. I'll take a look at it.

It's ok

Just was wondering if you guys had actually noticed the new version, since nobody had even really mentioned it. Of course if you don't have time to retest, that's ok, it's just that people keep talking about 1.14 and sometimes 1.15r, but few have seemed to notice the new and improved version.

A somewhat late reaction, I couldn't resist to make some remarks @NumLOCK.
There were some good points in that posting, here's just my €0,02

I think the approach with MPC (at least --q 5) is to have an accurate psymodel and to be efficient with the bits (low safety margin). As you state in point [6], this is a good approach.

Here you say: not all extra bits are useless, but most are (law of diminishing returns?)

What the quality settings (profiles) do is set the thresholds. With the same analysis that can mean that a value is above the threshold instead of under it. So it will be treated accordingly. A slightly higher --quality can soften some (hypothetical) artifact and, if so, it is likely to become inaudible at an even higher setting. This will be at a (bit rate) cost, also for the non-problem parts. Theoretical problems that have nothing to do with thresholds would stay.

The point is not if it is annoying or good enough (subjective) but IF a difference can be heard (more objective, ABX gives proof and credibility).

Then -why not- -q 7 or 8 ?
It is "the right thing"™ to stress the fact that MPC --standard is aiming for transparency in the vast majority of cases for the vast majority of listeners. It is on the other hand not a sin to use a higher setting, it is just silly if one did not prove to himself that the difference can be heard.

quotes edited by me
--
Ge Someone
What is this guy saying

Precisely. Above a certain threhold, only the underestimated parts of the sound can possibly be improved.


Not necessarily. the point of this codec is to suit your needs, whatever they might be


Yes and no. If you take the psychoacoustic definition of "problems", then no artifact can ever be 100% solved by increasing the bitrate. This is because the artifacting part is not priviledged with enough of the new bits. So, the solution is to fix the problem

My view is that you will probably always be able to find some sample that fouls a lossy encoder if you spend months looking and adding permenent dimples in your head from your headphones. However, I'm not going to waste hundreds of megabytes on songs that encode fine at -q5 for that one, half-second sample that you can ABX up to -q8. It just seems like such a waste of space.

You interprete wrongly: --braindead is better than --standard, but less efficient ('not as tuned')