Hi, since a couple of years I have been a regular reader of these forums, and before that the legendary "r3mix". I've always followed your recommendations, and because of that been a very happy MP3-listener. Thanks alot!

Up til now I have only listened to MP3:s at home, encoded with LAME 3.90.3 using --alt-preset standard. To my ears this has produced a completely transparent sound. I have also tried to participate in a 128 kbps-test by a Swedish HiFi-magazine, which proved to be extremely difficult, and I could hardly hear any differences at all. So let's just say I guess I don't have "golden ears" or enough training.

Anyway...yesterday I bought a portable MP3-player (256 MB memory) and I figured I should try to encode my MP3:s at a lower bitrate. 128 kbps. I checked out rjamorim's latest test and was surprised by the great score LAME received with "-V5 --athaa-sensitivity 1" so I tried it (with LAME 3.96.1)

It sounded really bad! The song I encoded was "Everlast - Black Jesus". The intro consists of guitar picking and here the MP3 sounds distorted. I don't know the correct translation, but the kind of distortion you get when you record a sound with too low sampling rate without lowpassing first. (In Swedish called "folding distortion" because the higher freq. are "folded" down in to the hearable spectrum.)

I didn't do any proper ABX-testing (sorry) but listened to the original, encoded with aps, and encoded with --alt-preset 128, using LAME 3.90.3. and none of those sounded like the one encoded with 3.96.1 and the above mentioned switch.

Have I done something wrong?
Is it better to use 3.90.3 with --alt-preset 128?
Is the "error" in my head?

//ix


PS: Tried to search the forum for a similar topic...but couldn't find any.

maybe you could upload a flac sample of the first 30 sec or so of this track?

I thought of that, but didn't know if it was legal or not.

Here's the sample
Everlast - Black Jesus
Hope I don't break any rules/laws now.

Samples up to 30secs in lenght are theoretically OK.

I think that it doesn't sound "really bad" for a 128 kbps encoding. That said, the artifact i can hear is unstable/distorted guitar attacks (not only preecho), i don't know if with "folding distortion" you're referring to the same problem.

If you try your sample with plain -V5 the artifacts are more evident. The "athaa-sensitivity 1" switch was proved to reduce artifacts with -V5 (and -V4 too) with those sample that manifested ringing/HF problems but, as stated in the past, -V5 --athaa-sensitivity 1 does not eliminate the problem but reduces it without bitrate bloating.


IMHO 3.90.3 --ap 128 is superior in some situations where ringing is still present with the vbr preset. During the test done to decide the LAME contender for the 128 kbps test i've rated 3.90.3 ap128 and -V5 --athaa-sensitivity 1 quite equally but the others have significantly preferred the vbr preset. See here: http://www.hydrogenaudio.org/forums/index....topic=21079&hl=

the folding distortion is known as aliasing - one frequency appearing as another (is that where the term comes from? i'm guessing here). i don't hear any of that here.

i'm hearing some of the guitar's harmonics coming and going intermittently, causing a kind of "watery" sound. not much pre-echo to my ears - it seems pretty well handled.

i think there's a bit too much quantization going on in the upper-mid bands (sounds like about 5-6Khz to me, but i'm just earballing it).

the bitrate is coming out quite low, so you might as well use --alt-preset 128 for this sample.

[edit]

i can't seem to ABX it at --ap128... but it comes out at about 138kbps though.

I successfully ABXed at the -V5 setting and at -V4, which is --preset medium. -V2 (preset standard) eluded me, however.

Hi again everyone, thanks for your replies.

I have conducted further, completely subjective, listening tests and have concluded that the problem is caused by the V-presets. Normal ABR-128 encoding in 3.96.1 sounds OK. As you've pointed out, the distorsion I was hearing was not aliasing, but the "watery" effect you describe.

Tried encoding with -V7 which sounded horrible. With -V5 it's a bit better, but still those "watery" effects I was complaining about in my original post. -V4 though, sounds OK to me...the water-phenomena seems to be gone.

I guess the V-presets haven't been fully developed yet?
Anyway, I'm really surprised that rjamorim decided to use -V5 for testing, and that it scored so high, when the artifacts are so obvious even to my untrained ears. ABR-128 seems (basing this asumption only on the Everlast sample) to be a clearly better choice.

Any explanation?

It's not the general case, that V5 is worse than abr 128. Of course there might be samples which sound worse in VBR mode, but overall V5 wins.

The explanation is embedded in your post: you're basing all this on only the Everlast sample. rjamorim based his decision on the results from multiple people testing a dozen different samples.

ff123

OK, I guess you know what you're talking about. You normally do!

Amazing though that I happened to stumble upon a problem piece the first and only track I encoded with -V5 --athaa-sensitivity 1.

Thanks for your patience.
I'll leave the flac sample online for some time, in case someone else is interested in testing it.