hi, i've now experimented with lame 3.93.1 and the recommended 3.90.3-modified encoder, with all sorts of flags, in an attempt to get them sounding better than fraunhofer's mp3 encoder, and cannot do so. a couple of weeks ago i did some abx test with "ABC/HR" on some orchestral records recorded in the late 1960s and was able to discern the fhg version from the lame version 12 out of 13 times. regrettably i dont still have the results so i decided to recreate the test using different source material to also provide some diversification in that the results are easily reproducable with a variety of sources, both analog and digital.

the vinyl was a recording of the new york philharmonic, besides the talented instrumentalists, they clearly they had some nice gear in 1968 and it shows in the quality of this recording rivaling and surpassing much of what is being released 3 and a half decades later. it was played back with an Ortofon cartridge, into an Echo Mia, at 24bit96khz, then downsampled and dithered to 16/44.1 before encoding. 3.93.1 --alt-preset-extreme, and fhg cbr joint stereo 192 (not sure of version, but within wavelab 4.02 so presumably quite up to date).

the quality of the fraunhofer encode of this vinyl record is virtually indiscernable from the original. the abx results here indicated guesswork, even going back and forth several times it was splitting hairs - close your eyes, and you can practically reach out and touch the instrumentalists. with the lame encode, subtle nuances are harder to notice, lost in a sea of harmonics. in areas where a lot is going on harmonically..which is quite often in an orchestra with dozens of instruments, the telltale 'edge' of the original recordings where an individual violinist jumped out from crowd was lost in favor of a mushy mass where everything congealed into a single 'sound' of slight dissonance. compared with the slightly distorted/damaged harmonics of the lame encode, the fhg version actually sounded quite pleasing to the ears and brought a smile to my face.. for modern music where theres just a bass, a snare, and some guy chatting over top, lame might do the trick but it simply doesnt seem to be up to the task of faithfully recreating an entire orchestra.

since there seemed to be many response to my prior post regarding a lack of proper protocol for such opinions, and since i'm quite confident that the results could be easily reproduced, i did this test again, this time not with analog material but with boymerang's "balance of the force" cd. not created in an orchestra hall but in a studio with loads of digital fx etc. the extreme studio skill of the producer creates a sense of 3d space which would be good candidate for encoder destruction. lame 3.90.3-modified was used, --alt-preset extreme -k to make sure stuff above 19khz was not filtered and fhg joint stereo cbr 192 again.

ABC/HR Version 0.9b, 30 August 2002
Testname: lame fhg
1L = C:\sounds\boymerang\06fhg.wav
2R = C:\sounds\boymerang\06lame.wav
ABX Results:
Original vs C:\sounds\boymerang\06fhg.wav
4 out of 5, pval = 0.188
Original vs C:\sounds\boymerang\06lame.wav
5 out of 5, pval = 0.031
C:\sounds\boymerang\06fhg.wav vs C:\sounds\boymerang\06lame.wav
14 out of 16, pval = 0.002

discerning the lame version from the original, it's like a veil has been unlifted off the snare drums. fraunhofer is a bit more challenging, you definitely have to focus while listening, and then it has only marginally less presence than the original. given the size tradeoff i think it is a worthy compromise. it should be noted that the lame file was 14 mb while the fhg file 11.1. what it was doing with those extra bits im not exactly sure. except for a couple spots where the sound consisted of nothing but a dull atmospheric drone, discerning between the two was quite simple, the fraunhofer simply feels like its in the room with you, while the lame, playing just as loudly, feels like its in the background. fft later revealed slightly more loss of high frequency information with the lame version which could account for some of the dulling, but the fatigue and 'im not listening anymore' aspects of the lame encode are proably more attributed to the math which consolidates the harmonics above 5 khz or so, pure speculation might suggest fhg achieves a more pleasing effect by reproducing this region in a more logarithmically symmetric fashion, since distortions and deviations in this region can cause your ear to tune the sounds out, secrete earwax, etc.

so for me the choice is clear in terms of filesize and quality. i can wholeheartedly "Recommend" fraunhofer's encoder to anyone looking for a high quality reproduction of their existing records or cd's. the lack in ubiquity of available players rules out formats such as ogg and mpc, although if someone comes up with a portable player that uses the .ape format (does one exist yet) i can't wait.

I think I'll search out for a nice and warm place and take a look at you being incinerated by the community ...

By the way ... you have proven only one fact ... that you can clearly ABX the Fraunhofer encoder (at CBR with faulty JS) over LAME (at APX with questionable settin -k, that is) with given material. The subjective listening effects you describe might be sensed/judged different by different people ...

EDIT: exchanged some words

Sigh

GOD DAMN IT CARMAN! - Read what I wrote here - http://www.hydrogenaudio.org/forums/index....topic=14469&hl= - adding -k to the presets will DECREASE QUALITY! Pretty graphs do not mean good quality. But instead of listening to me, you start another post stating the same mis-informed thing. Just try to abx a straight --alt-preset extreme command line without the -k. I bet you will be supprised...

edit - If you don't believe me that graphs are a bad way to judge quality, check out this command line (without quotes) - "-V0 -q9 -b256 -B256 --resample 48 --interch 1 -md -p --noshort --notemp --nores -k --strictly-enforce-ISO --nspsytune --athlower -56 --ns-bass 2 --ns-alto 12 --ns-treble 9" - it MUST sound great, look at the pretty graph

Can you explain why you punish lame with the -k switch but let the fhg encoder filter the high frequencies? Next time use only --alt-preset standard.

Oh, providing the samples would be nice too

Have you tested this on one or many samples? If only this one, it might be an isolated hickup by the lame encoder - then it would be interesting to have a copy of the test case you used. If it's consistent throughout many tests you have done with different music it could be that you are more sensitive to the flaws in lame apx than in fhg... That would be rather surprising since most other people (at least around here) pretty much agree that lame aps/x sounds better than fhg@192.

@carmen: Simple concept. You forced LAME to use some bits to encode frequencies you'll never hear (unless you're a bat) instead of letting it use all of those wasted bits to instead encode the portion of music you can hear, hence decreased sound quality. More bits used above 19KHz means (generally) less bits the encoder uses in the audible range. And no, most people cannot hear anything (consciously) above 19KHz, and many (like me) cannot hear anything above 17KHz even.

Like everyone else is saying...use the --alt-presets by themselves, and don't cripple your encoder by trying to "tweak" it. Not that your intentions are not good, but it's just that Dibrom and Gabriel and others who developed/tested LAME spent much effort on tweaking the --alt-presets to make them more effective than what just about anyone else could do manually at the command line. AFAIK, no one has ever proven any significant sound quality problem with the --alt-presets when used by themselves for normal music. The exception is perhaps on problem samples, but almost any encoder setting can be broken by a particular problem sample (to some people's ears). Psychoacoustic audio compression is not completely perfect, but the accepted presets/quality settings are accepted around here for a reason...because they almost always work better than "tweaked" command line parameters.

Carmen: Can you provide some example clips for download for others to verify your findings. Also, you might want to do some blind rating with ABC/HR. ABX itself can just tell that you are able to hear a difference.

@carmen: I can say with 99% confidence that you are not able to hear the 20kHz part of music. This means that by using the -k switch, you reduced the encoding quality.

Well this looks like a case of user error, however I'd still like a crack at those samples. Carmen could regardless be onto something.

Could you upload those files ?

Heres a good place: http://www.hydrogenaudio.org/forums/index.php?act=SF&s=&f=35