Greetings!

Yes, I've been using the search function, but people don't seem to love this preset very much. Apart from some unclear statements, some ABX tests with weird results and obscure recommendations not to use this preset, there is not much publicity around it... at least from a user's point of view.

AFAIK, lossy audio compression is intended for achieving a dramatic decrease in file size without losing much of the audible sound. APS has become a kind of standart (just what it says) of LAME compression, but it still produces files around ~200kbps. Although I'm sure some of you gold-eared people can ABX aps from the original in 100% cases, let's admit that it's a bit of an overkill for the average listener. After all, lossy audio compression loses sense gradually by every kbps increase of a particular encoding. Especially when it comes to portables.

It seems, an average bitrate of about 160 should provide transparency for most people. Not that it mattered much, but in the LAME html documentation the --preset medium is, if memory serves, the smallest compression mentionig "transparency" as such. Theoretically, this bitrate range has the potential to be the best choice, the best size/quality tradeoff for most people. Judging by the quite fine result a "-b 160 -m j -h" delivers with v3.90.3 (give it a try for ol' times'sake! ), and the fact that CBR 192 has been a standard in the scene for years, VBR averaging around 160kbps SHOULD be a safe choice for most of the (ear-wise) mortal people. I suspect the APM preset was developed with something like this in mind. Now, why is it used so rarely? I'll try to answer the question myself, and hope you help me with it.

In my personal listening experience, I find v3.90.3 --preset medium fine. But 2 things discourage me from using it without worry:

1) the recommendations from other people are usually negative, like:
"don't use it" (-no comments);
"use aps / aps -Y / --preset 160 instead" (-too big / why then bother to include APM?);
"it hasn't got code-level tweaks";
"it sounds muddy"(-?);
"it hasn't been tested thoroughly" (-sure, you can't test it without using it);
"VBR at <160kbps is unsafe" (-shouldn't VBR ALWAYS sound better than CBR? );
"APM was designed for newer versions of LAME".

2) after performing the "Microsoft-experiment" (as I like to think of it ), i.e. inverting the original and mixing it with the encoded sound, the remaining difference is occasionally quite loud and includes some weird mid-frequency "warbling", which means that there is relatively much sample-level difference to the original. (Later I discovered that the same warbling was present in 3.90.3 APS, though less. There is almost none in CBR 160). Of course, this doesn't neccessarily mean that you hear it when listening to the music, but still - the difference is there.

I found an ABX result at HA where APM v3.90.3 and 3.95.1 were compared, and the verdict was clearly in favour of 3.95.1. Indeed, after performing the experiment mentioned above with v3.95.1, there were much less artefacts left after the cancelling-out. The bitrate was somewhat bigger, but not that much, and a whole bunch of APM settings seems to have been changed from v3.93.1 to 3.94 and up. So I figured the 3.95.1 APM must be better and switched to it, until I stumbled upon a passage in a track which sounds rather bad with it. I tried the same with APM 3.90.3 and the result was much better @ a smaller bitrate!

So I guess my questions are these: is the preset adequate for usage? Is the quality proportional to the bitrate? If so, which LAME version to use with it? How does the future of this preset look like (it has been changed much with 3.94, but not neccessarily to the better)?

Thanks,

Immo

I will add to this Q. myself... I've been trying to read up on best preset for the use transcoding from MPC --xtreme to MP3's for portables, but no two persons here at HA seems to agree with each other (hardly even with themselves), so one get's a bit con-fu-se...

Some say --alt-preset standard -Y, some say go with ABR, like --alt-preset 144 (insert target bitrate), and a few mentions medium. (BTW: should it be --alt-preset medium or --preset medium ...?)
Personally I think --alt-preset standard -Y must be bitrate overkill for this particular use (?). Maybe even medium is?
But using SEARCH here at HA don't clarify much, only adds to the confusion....

???

Really?? I believe that would be news to quite a few people, myself included. Is that the complete test, or just one sample from a larger test? Got a link to that test result?

All of the ABX test results I saw indicated there was not much difference between the two, and that in some cases 3.95.1 actually had some issuses that 3.90.3 did not have.

i use lame 3.95 --preset fast medium on my portable and from my personal listening tests (yes, i do abx...) it sound *much* better than e.g. --preset 160. i don't see any reasons no to use it.

regards; ilikedirt

woops: wrong account... but i'm ilikedirt indeed

I've been encoding for my portable in --alt-preset medium; probably because of normally being outside I haven't noticed any problems what so ever.

In my case, anything higher would be an overkill, killing battery life of my portable slightly quicker and taking up more disk space.

Rob

hmm, as for why so few use APM here.... i'll take a risky guess(i may be wrong):

1. People tend to encode higher than necessary.... they're used to it. Like "i will probably never hear it, but i just want to be absolutely sure".

2. I guess for many people in the past 192kbps CBR was the norm. However, their feeling tells them to continue at that bitrate by using APS, and take advantage of higher quality. Even many people who encode with the presets seem to subconsciously still be afraid about using a lower average bitrate, even if common sense says that its unnecessary.

3. Herd-Instinct..... way more people use APS than APM.... so newbies also learn to use the setting which most others use.

I'll mention a few answers based on what I have seen spread around this frum on the subject... took me quite a while to find so I figure I'll save you some time I've already spent.

Here are a few of the main things that drive recommendations when it comes to mid-bitrate LAME encoding:

1. MP3 in general, due to design limitations, can be very inefficient when encoding frequency content above 16 kHz (I don't remember the specific technical reasons why, but searches for "sfb21" in the forum should allow you to find them). Any attempts at achieving true transparency with MP3 hit the bitrate bloat that comes up as a result. This drives people's recommendations at modifying --alt-preset standard with the -Y switch (a switch that tells the encoder to ignore encoding any frequency content above 16 kHz that would contribute to the high frequency bloat) or with reduction of the lowpass cutoff frequency.

2. Limitations in the psychoacoustic models used by LAME can cause it to improperly use too low a bitrate with some content, so the LAME aps switch by default does not encode frames at less than 128 kbps unless digital silence is involved. This is enough of a problem that at bitrates of 128 kbps, even CBR can be a better choice than VBR. ABR methods tend to be better choices than either CBR or VBR at bitrates below 160 kbps. Reducing the minimum bitrate used on the standard preset is *not* recommended as a way to reduce average bitrate--it's not very effective unless you go down low enough that obvious sound artifcats will crop up in encoded files. This is the reasoning behind people's recommendations for --alt-preset xxx (where xxx is below 160).

3. Different people have different sensitivity to lack of high-frequency content and presence of compression artifacts. The aps -Y method of saving bits focuses on removing certain high frequency content, while medium VBR preset seeks to save bits by allowing some compression artifacts while keeping more high frequency content. No one really recommends medium consistently yet because it doesn't stand out often enough from alternatives. This may change in the future as more work is done, though.

4. aps is meant for transparency for almost everyone, with the lowest bitrate that will not sacrifice quality *at all*. Its deficiencies are usually a result of limitations in LAME that can only be resolved with considerable changes to the psymodel, or unavoidable limitations to the MP3 format itself. LAME was originally developed with this purpose in mind. The fact that it *is* almost always transparent is confirmed by the massive amount of testing that took place, much of which is documented in this forum, so people rightfully trust it. It is *not* the best choice for portable use, due to the noisy envirnment portables are used in--transparency is wasted there, and all you have left is a bloatd bitrate. Hence the effort for new presets at more "useful" bitrates. The medium preset shifts the priority from transparency to best quality at a bitrate between 140 - 180. This causes problems in development, because transparency is a very clear goal, while the definition of "good enough" changes from person to person. Also, medium's target bitrates seem to push the limits of the LAME psymodels, making it difficult to get its quality head-and-shoulders above similar-bitrate options.

Hopefully this information is helpful.

I've been quite the lurker here (I already spend plenty time typing on other car/tech/game boards), but I just thought I'd add my story here.

In about 1997, I thought the sad 128kbps MP3's people were making with that awful Creative Jukebox or whatever software sounded great.

When I got a Rio 500 in 1999, I thought using FhG at 160kbps would be nearly perfect archival quality. I still use the Rio all the time, and 80kbps is more than good enough for me when I'm running with tiny little headphones.

When the fancy --r3mix thingy was added to LAME 3.89 in 2001, I thought I was getting CD quality for sure. I somewhat reluctantly switched to --alt-preset standard in LAME 3.92 in 2002, seeing as r3mix is smaller and sounds good.

In 2003 I got a car MP3 player, and I actually re-encoded a few things in --alt-preset 128 just to fit more songs on a disc. Downloaded 128kbps singles sounded so-so, but everything I'd done myself sounded perfect.

Then, last month, I put a set of 4 Infinity Kappa's in my car, and BAM!!! Suddenly everything sounds horrible except for my --alt-preset standard stuff.

So, transparent is all relative. To me, 80kbps is transparent on headphones during a workout... But having good tweeters will change your mind quickly. I'm considering just going with --alt-preset extreme from now on.

--alt-preset standard is about 85MB per hour, and extreme is only about 110MB per hour, so I'm going to see if this helps the high ranges and probably switch.

Anyway, I mention all this because I think it's the type of progression many people go through. It doesn't stop there -- look at how many people use Monkey Audio and FLAC. This is why most people would suggest strongly that you err on the side of quality. You can get a 120G hard drive delievered for $74. You can get reliable DVD-R media for $1. There's no reason not to go for overkill. If you use a solid state portable, you can quickly transcode albums down to really low bitrates with the -fast switch. Don't try to come up with a compromise that is good for portables and for archival, because a year from now you'll realize you were wasting your time.

maybe you should try LAME 3.95.1 ! It gives me much lower bitrates than any other LAME release. I use --alt preset standard and it gives me a bitrate around 200kbps.

Indybrett:
If you look up

http://www.hydrogenaudio.org/forums/index....pic=17499&st=0&

and search for "medium", you will find a couple of results that indicate 3.95.1. --preset medium superiority. As I said, my experience is the opposite.

Powerpigg:


It seems that --preset medium in version 3.94 and up matches this description, whereas previous versions tended to cut down on hi-frequency content. From that point of view I don't like the development, because I don't hear above 16kHz. The up-to-18kHz "speckles" you can see in a spectral image of an --preset medium encode serve more as a psychological comfort , and in case other people (who DO hear more) listen to it.

Schimpi (Ilikedirt):


May be, but have you checked if it sounds better than --preset 160 in 3.90.3?

Dimension:


You've got a point there. I had a similar experience when I bought a couple of Sennheiser HD202 headphones (low end, I know ). Treble-heavy as they are, all the Internet-128s suddenly sounded horrible with them. --preset medium sounds OK. But I might want to come up with a compromise for portable and archival, because I use the headphones both at home and outdoors, and since they are the most mp3-sensitive thing I've got, and are closed and isolate pretty good, my demands for archival and portable are quite similar.

Some questions still left unclear, I think I'll stick with 3.90.3 --preset medium for a while. I wish I could upload the sample where 3.90.3 beat the crap out of 3.95.1, but my Internet presence is very limited, sorry folks Maybe some other time.

Immo