Hello, I've found a dreadful test sample : Autechre - Anvil Vapre.
Since I don't know all the test sample you used, I apologize if it is already well known (especially with Dibrom enjoying this kind of music, if I'm not mistaken).

The dropouts are even worse with alt preset extreme at 253 kbps than with alt preset standard at 232 kbps ! They are still obvious with alt preset insane.

The full passage can be interesting because the sequence is repeating with differences, so that the dropout is more or less present, but it's 7 MB in lpac.
I can make a shorter version with the worse dropouts.
You can send me a private message, ff123, so that I can send you the sample in order to put it on your webpage.

By the way, there was already a sample from Autechre, wasn't there ?
Should we award them the status of "most difficult band to encode" ?

Sorry, the title of the track is actually "Second bad vilbel". "Anvil vapre" is the title of the CD.

This is actually from one of their albums that I don't own (along with Cavity Job, We R Are Why/ Are Y Are We, and Radiomix.. which are, of course, much harder to find...)



Hrmm... interesting.



7 Megs is fine for me at least.



There are actually quite a large handful of samples from them that I've used in testing before. Some other bands from this genre also have some very difficult music, like Two Lone Swordsmen for example.. numerious samples from them and Autechre were used to help tune Vorbis even.. with Garf's Tuned modes and Monty's much-improved-as-of-late pre-echo handling stuff.

To be honest, I don't even really bother encoding this kind of stuff with MP3 anymore because it's impossible to get it to sound good IMO. If there aren't other artifacts (like maybe this dropout stuff), there's lots of pre and post echo due to the nature of the music. In fact, this kind of music is what originally drove me to use MPC...

Anyone ever tried Jean Michel Jarre's Equinoxe and/or Oxygene? I swear I can hear dropouts and other obvious artifacting with --alt-preset insane, although I haven't ABX'ed it -- perhaps I will one of these days.

Fascinating music. I discovered them via a collaboration with David Sylvian and is at present quite intrigued by LP5 and Tri Repetae+++. After reading about them on Allmusic I also sampled music by Merzbow.That was too "avantgarde" for my taste

Garf: Shouldn't the Anvil Vapre and Garbage EP's be included when you buy the american release of Tri Repetae+++?

A bit OT:
Yeah... LP5 and Tri Repetae are great. Have you heard Confield from Autechre yet? It's one of their best releases IMO. For those interested, you can actually read my review of the album here.



Ah, yes you are right. I totally forgot about this... so n/m, I actually do have this track I'll take a listen shortly.

I just examined this track in cool edit and took a quick listen, and I'm almost positive that this is the exact same problem seen in the dogwhistle clip which also gives --aps problems. I haven't actually encoded and listened yet but I will soon.

The problem though (I think) is the presence of a very strong tone over 16khz and incorrectly calculated masking from such. In the case of the dogwhistle clip, this is actually a fault of vbr-old... vbr-mtrh doesn't seem to have the problem. I think part of the fault is related to an inadequate spreading function but I'm not positive.

At any rate, I have found a way to fix the dogwhistle clip (and possibly this one), but implementing the fix isn't easy because it also has the side effect of bloating the bitrate in other cases. Maybe there's a better way to implement this but I'll have to experiment more...

In fact, this sample is a terrible melt of pre echo (during the scratches), dropouts (during the scratches reverb decay) and ringing (just after the dropouts).

To remove the 18185 Hz tone actually helps (low pass filtering : 16 290 to 18 101 Hz, -60 db), but all the problems are still there, reduced.

The sample is now in FF132's hands. It will be soon online.

sample is up as badvilbel.pac at http://ff123.net/samples.html

I must say that file makes a hell of a racket.

ff123

I have to ask... Is that .pac the original or the sample run through the encoder?

Hehe, is somebody seriously listening this kind of music and enjoying it?
Yeah, it's quite typical pure "effect" clip.

Btw. ff123, I just sent you PM.

OT



Heh. Yes, definitely. Some types of music have dared to go beyond the traditional Western musical senses of melody, harmony, and rhythm that we are conditioned to like. This kind of music may be hard to listen to at first because it lacks the traditional identifying points that many people are used to holding onto. But once we are able look, or better yet listen, beyond what conditioning has taught us to listen for in music, many new possibilities open up. Some of it may be harsh, frightening, and even beautiful but the most common goal of experimental music is to move beyond what has already been done in the past. In fact there is no doubt that experimental music of the past has even had a direct influence on the mainstream music that we are now accustomed to.

Well enough of that already

Some of my favorite experimental/Modern groups/composers: Aphex Twin, Autechre, David Tudor, John Cage, Plaid, Squarepusher, Steve Reich, Karlheinz Stockhausen, The Future Sound Of London, La Monte Young.

That is one of my favorite genera. I have heard Autechre, but there is no place here that sells their albums. I have seen a couple on the net, but I have not had the money to spend for a while. As stated to most people the stuff can often sound quite awful. Of course most the stuff people listen to today I would not even use their CD's as coasters. The best I guess you could equate it to is moddern day classical music. The fun and enjoyment comes from listening and picking the songs apart and listening for the interlaced rhythms and effects. Course you can't listen to it all the time. So in the meanwhile I listen to my classic rock, trance, gabba, acid, house, ambient, goa, rave, and a little nonverbal turntableism thrown in.

On an OT note, somehow I can't stand Autechre, and I even like noise "music" (Merzbow has some interesting stuff, and Haus Arafna is amazingly excellent). Something about high-pitched squeaking and scratching and clicking that really grates on my nerves. Same with Aphex Twin.

On a moderately more on-topic note, that has to be the worst-performing clip I've ever tried to encode. Perhaps Autechre use LAME to tune their music, but in the opposite way...

Yeah.. this is why I say that MP3 just isn't a good choice no matter what for this type of music. Even if you can eliminate most of the dropouts, you're never going to be able to elminate pre-echo just due to the limitations of the format.

Beyond that, I don't actually think it's possible to tune away most of the other faults LAME produces when encoding this material, pre-echo aside. It might be possible within the constraints of the format, but the psymodel used in LAME is not advanced enough IMO to do this properly, and I'm not sure that it will ever really become so.

I would like to improve the performance of --alt-preset standard on these types of samples, but I'm not really expecting much more progress at the moment. Certainly I wouldn't want to tune for these cases in detriment to more "standard" musical signals.



Hrmm.. seems that the tone may still be effecting masking calculations if the lowpass comes that close to the tone. Actually it may still be regardless of whether it is lowpassed or not, I'm not entirely sure how the psymodel works in that regards since I haven't examined the lowpass stuff in some time.

Maybe nspsytune2 will help...

Probably some, but even with nspsytune2, unless Naoki has made many changes beyond his original plan, I don't think it will quite reach the level I'm thinking of. If I understand it correctly, the main goal of nspsytune2 is to allow for better tonality estimation..

There are still a lot of other little areas though in the psymodel which were tuned a very long time ago and under what we'd probably now consider questionable circumstances. The spreading function is an example for one.. as is the block switching -- try encoding drone_clip (another autechre clip) with LAME to see what I mean.

At one point, Alexander was actually working on implementing a non-linear psymodel which I think would have been probably the best bet, but development dropped off sharply after the very first addition to the codebase. I'm not sure if anyone is continuing that work at this point or not.

It would be interesting to see what could come about from working the benefits that may be offered from nspsytune2 into the benefits that may be offered by a non-linear psymodel. I don't think Naoki feels this non-linear model will help much though (even though some very advanced codecs like MPC use ideas from this I believe).

I'm sorry. You forgot to mention Ulver... *blam*

Northpack

Pio2001 sent me the full sample, and I have made it available (both LPAC and FLAC) here:


http://personal.www.umich.edu/~aklenk/music/vilbel.flac

http://personal.www.umich.edu/~aklenk/music/vilbel.pac

This is quite the problematic clip. I must say that i've never really heard music like this before, though i kind of like it! Music comes in many forms, right?

--Andrew