Hi for all;

I encoded some CDs for the portable player of my brother.
I used Lame 3.96.1 --preset standard for this. The sound quality of the files are excellent. But; one particular song produced a strange and awful artifact; around first 2.5 seconds of the song.

This artifact sounds awful. I checked with -aps and -ap extreme with 3.90.3 and 3.96.1 with the same results. The rest of the songs sounds great. I think that is important to notify this. I uploaded the sample here:

Note: RIGHT CLICK and SAVE TARGET AS...

Download

This sample is only the first 5 seconds of the song.

Thanks a lot in advance and feedback

Can you hear the artifact only on the portable or on the pc ? I don't hear anything too weird let alone abx. -ap extreme should have reduced the artifact - if not I suspect some other problem.

The artifact is noticeable very easy with my headphones in my Stereo system. I obtained easy ABX without a wrong trial. I am sure that is not the PC

I can't hear any artifact using 3.96.1 --preset standard, but maybe my hearing is good enough. Definitely nothing that grabs my attention by slapping me in the face.

Holy s**t - I got - its around 1.0-2.2 - very bad noise

Easy abx 8/8 - on aps. Complete psymodel failure.

This is strange... for my ears is very easy, noticeable and awful

ok, good.. Thank You for confirm this. .

I guess I should feel fortunate I can't hear it then. I tried ABXing but only got 3/8 so entirely chance.

What kind of noise are you guys hearing?

Shadowking, please check with --preset extreme

It's a strange dropout with phasing; first time that i heared something as this

abx 15/16 on -ape, -api is too difficult.

OK, thank you a lot.. I didn't check with -api, only with -aps and -ape. I go to check...

With -api is very difficult to me, too

3.90.3 -Aps seems ok.

I checked with 3.90.3 -aps and sounds awful too

But; in -api with 3.96.1 is very difficult to me

A few thoughts: We should have been able to abx -api, but maybe cbr helps here because the vbr model is not giving enough bits ?

You might want to try 3.96.1 cbr 192k or even --preset fast standard ...

Hmmm.. interesting.. I go to check too

bad artifact again with --preset cbr 192 ... [Lame 3.96.1]

With --preset fast standard, the same bad thing.

with lame 3.90.3 in --alt-preset cbr 192 and --alt-preset fast standard the same bad thing again.

My god!! this thing sounds awful !!!

I can ABX without a wrong trial too, in:

-aps, -ape, --preset cbr 192.

with -api is difficult to me too.

NOTE: tested with 3.96.1

just got 8/8 with 3.97 alpha 6 -V 2. not a disturbing noise to my ears, barely noticable except when abxing.

i'm not all that used to ABX in general (my headphones are always off getting the headband fixed, you know...?), but even when i felt unsure i seemed to click the correct samples.

will try with insane (i hear extreme isn't really much of a cushion, and if it fail with standard it's likely it'll fail with extreme).

[edit]

i didn't stand a chance with --preset insane... will try extreme now

[edit 2]

extreme was just as easy as standard. interesting. must be a vbr thing.

Maybe... could someone explain it?

or maybe... a possible solution? (for example a command line to --preset standard or something similar?).

Thanks a lot for everyone for your replies

Thank you for confirm this weird thing...

-api for me is very difficult...


I believe the same thing

What will be the possible solution??

For the record, I still can't hearing but since all of you can my hearing just must suck

Since you guys can do it with the VBR modes but not 320k CBR then it must just be something with the VBR mode. That or throwing 320 frames at everything masks the problem to the point you can't hear it.

It's pretty interesting to me even though I can't hear it and have failed trying to ABX it several times.

8/8 on -alt preset extreme

first abx ive done, not too hard to tell, like a warbling/phasing problem.

dreamtactics:

try concentrating REAL hard on the bell tingling as it recedes. the encoded version to my ear almost ends up a tone lower. the phasing i wasn't so good at pinning down, but i "felt" like i heard it. subtle differences are real hard to describe, because you're never sure you're not imagining it. if you lose concentration even for a second you can completely lose faith in what you were hearing and have to go back to listening to a and b for a while.

i need more practice ABX'ing, because in theory my hearing is pretty good (which is a good thing cause my other 4 senses are pretty average - i've got almost no sense of smell left thanks to years of turps-fumes )

[edit] btw, i did this on nice cans and with the volume at slightly above the "comfortable" volume. both my phones and my ears are good up to 19.8Khz, but my sound card motorboats a fair bit (it shuts up when i'm doing an Xvid encode!!)

can confirm @ aps w/ 3.96.1. If you concentrate on the little bells, you should spot it easily.

EDIT: MugFunky beat me to it

@DreamTactix

For me in particular, is difficult to hear certain artifacts without headphones.
Try with headphones

I couldn't hear it first with aps (3.96.1), but then I tried with 128kbps and after a while I was able to identify what to listen for. Abxed 8/8 at 128kbps, then tried again with aps; 8/8.

I don't find it annoying and wouldn't even have noticed it if just listening casually.

I was using headphones. It's most likely just my lack of training when relating to some artifacts like phasing. I can spot easy ones like ringing and pre-echo (finally )

Listen to the bells. There is a second where the bells peak in amplitude causing a fluttering / dropout artifact.

Many thanks to all for your replies .

I am very impressed with the fact that no one of the LAME community has not answered to anyone of these commentaries.

At work with my crappy integrated sound and Etymotic ER-4P headphones, I am able to ABX 8/8 with aps. I'm not really sure I'm hearing the artifact that you speak of; I could ABX based on the fact that in either the original or the encoded version the chimes seemed to die out faster (from 1.0-2.2s). I am unable to ABX at my usual setting of V1.

I might have to try again at home using my Grados and better soundcard.

Finally got it with LAME 3.96.1 --preset standard



Will try with --preset extreme in a few minutes. I missed one pretty early on but I could nail it easily after that. The chimes seem to end at about 1.2 seconds with --preset standard while they continue in the original until about 2.2 seconds. Really subtle but it is there.

You guys are really good to have spotted that because during casual listening I never would have.

Finally good...

""The chimes seem to end at about 1.2 seconds with --preset standard while they continue in the original until about 2.2 seconds""

Yes, and; in this time (1.2 to 2.2) phasing is very audible and ugly... causing dropout in the bells sounds...

ABXed 10/10 with 3.96.1 --preset extreme. --preset insane was very difficult for me as well and I couldn't ABX it. As I said the effect is really subtle but I finally do hear it.

For me --preset insane is very difficult too.

The strange thing is that the phasing effect for me is very disturbing and noticeable very casual without headphones, but; your ears are very good, for me is a little difficult hear pre-echo artifacts

Well until last month castanets was 100% transparent to me with a lot of settings. I got frustrated that I couldn't hear the pre-echo. Now it's not so bad. I'm not as trained as a lot of people here, but I'm learning.

hmm. perhaps it's something that occurs when the channels mix then. i didn't notice any phasing as such, i just noticed the chimes being audibly different (in a frustratingly hard to describe way). but i haven't heard this sample with speakers yet. i've noticed when messing with lame settings (don't try this at home), certain artefacts are more audible through speakers than headphones. i guess this is because of phase cancellation between left and right, which does not occur when the channels are isolated.

perhaps we could mix our source and our encode down to mono and see what happens?

Yes, this is strange...
My opinion is that this killer sample will be very good to the development the Lame (I agree with the fact that this artifact is very strange and particular, as awful too).
Considering the fact that Lame is a very good mp3 encoder as consequence of various tunings made by many exhaustive listening tests of Killer samples (and music files, of course); the silence of Lame community is IMO a very bad thing....

3.96.1 aps: 0/1, 10/15, 2/3, 8/8 (was able to zero in at the last trial)

My first semi-successful APS ABX. Yay. Equipment was ER-4S through a convoluted setup I just realized was pretty noisy.

Very, very hard to spot. I would not have noticed this unless it was pointed out to me.

This one is disturbing because lots of music has these bells. However, I think that if you abx other bell samples they will probably be clean.

I am sure they will look into it and probably have already. Also killer samples are not representative of a codecs quality. I have accepted that all transform codecs will fail with some music no matter how tuned they are, and the artifacts can be very annoying. Use lossless or hybrid lossy with sufficient bitrate if you don't want to hear these ugly artifacts. For portables and car stereo you might have to compromise, but for PC listening the answer is easy with today's storage.

Yes and Not...
Remember the fact that many killer samples are really music !!!
Castanets is a good example, the sample in this thread too...

and Lame is a very good encoder as consequence of many tunings with these difficult samples.. that.. many people uploads...

Many people not agree with this, I am one...
Why waste space if with a good encoder, with good tuning you got transparency??
I don't have any portable player, and, my Hard Disk and PC are old.

For example, a friend near me encoded all your music with vorbis 1.1RC1 in Q7 setting... My friend and I mades a lot of ABX tests with many of your music... the results were complete transparency. The average bitrate was 215 kbps, very small if you compare with lossless codecs..
In my own experience I obtained with Vorbis 1.1RC1, Q5 (ITP=4) total transparency too, and the average bitrate was around 180 to 190 kbps, very small too..

One detail: For killer samples in my own ABX tests Vorbis performed better than Lame... and, for Vorbis Q7 setting many problematics samples are very very hard to ABX (for me only with headphones...)

Could someone try--preset standard --noath setting?
I could also easily ear this artifact (aps and ape are in the same situation), but it seems to disapear with --noath switch (tested with 3.97a6).
--athlower is another solution.

Also tested with lower VBR setting (-V5...V3) and --noath: distortion is still audible and annoying.

I tested your settings; they performed very good...

With "aps --noath" the artifact is present, but very lower.. I believe that is aprox. only 30% than original aps... and I believe that for casual listening is impossible for to hear. I heared only as a little dropout (with headphones).

With "ape --noath" the situation is different... For me is transparent, and very very hard to ABX.

Here my ABX results:

3.97a6 --preset standard --noath [7/8]

3.97a6 --preset extreme --noath [3/8] Very Hard

3.97a6 -V 3 --noath [8/8] Easy ABX

3.96.1 --preset standard --noath [7/8]

3.96.1 --preset extreme --noath [2/8] Very Hard

3.96.1 --preset standard --athtype 5 [8/8] Easy ABX

This switch perform very good; the final results are a lot better than traditional settings..

Many thanks Guruboolez for your reply and professional help

NOTES: For some strange reason Lame 3.97a6 and 3.96.1 crashed with --athlower in my PC

IMPORTANT: I discovered in this moment the problem... this is a complete cutoff around 0.8seconds to 2.2 seconds for the frequencies range of: 9Khz to 11Khz !!
tomorrow I come back with the frequency analysis graps.

@Guruboolez

I discovered that the problem is in the range of 0.8 seconds and 2.2 seconds Lame (aps & ape) cutoff the frequencies around 9Khz to 11Khz aprox.

Look at the pictures:

Picture 1): Original WAV file [1.124 seconds position]

http://www.geocities.com/erasmo32/Orig_Killer.gif


Picture 2): Lame 3.90.3 --alt-preset standard [1.124 seconds position]

http://www.geocities.com/erasmo32/aps3903.gif


Picture 3): Lame 3.96.1 --preset standard [1.225 seconds position]

http://www.geocities.com/erasmo32/aps3961.gif


This cutoff [aprox.: 9Khz to 11Khz] behaviour is almost constant till aprox. 2.2 seconds... after of 2.2 seconds the phas¡ng effect decrease.

I listened with my headphones, with Cool Edit with Frequency analysis enabled [in real time], I am sure that for my ears the ugly phasing and dropout effect is consequence of this cutoff.

I could hear, clearly; the artifact in the moment that this cutoff was present in the analysis frequency window...

I could see that the channels (L & R) from time to time switching ON and OFF in this range of time: 0.8 to 2.2 seconds...

Many thanks to all who help me with your replies and support. I learned a lot with this experience .
Without your replies to me it would be impossible to decipher this..

Now... I have another doubt...

Maybe:
Will be fix this bug in new Lame versions??
Could be "dynamic ATH" a possible solution??

Many thanks and please any new feedback will be very welcome

PD.: Sorry my bad english

[Inlined images removed by moderation as they no longer serve their original purpose.]

This frequency gap isn't necessary a problem (though it's apparently the case for this sample). Many lossy encoders aren't encoding intermediate frequency region for specific samples (probably for masking reason), and sound is nevertheless transparent.

Yes, it's true. I could confirm this in my Cool Edit Pro, with frequency analysis window enabled. I listened very carefully with my headphones, and could hear as the ugly artifact appeared just in the moment that the cutoff (9khz to 11khz) appeared in the analysis window. This artifact disappeared of my ears just in aprox. 2.2 to 2.3 seconds, in the exact moment in that the cutoff (in the analysis window) disappeared too.
Very good GrayWolf



The same situation for me.

EDIT: the file analyzed was in aps lame 3.90.3