Skip to main content

Notice

Please note that most of the software linked on this forum is likely to be safe to use. If you are unsure, feel free to ask in the relevant topics, or send a private message to an administrator or moderator. To help curb the problems of false positives, or in the event that you do find actual malware, you can contribute through the article linked here.
Topic: Public Listening Test [2010], Item Selection (Read 48777 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

Public Listening Test [2010], Item Selection

Reply #50
That's a completely different case than (perceptual) compression of audio. Binary (non audio) data can have regularity (and also complexity) several magnitudes larger than usual audio data. Binary (non audio) data can have repetitive patterns spanning larger ranges than what could be found by brute force search. So it can make sense to tune a lossless data compressor for specifically structured patterns out of the infinite problem space. With the exception of looped, electronic music the same is not usually true for audio.


That's interesting; I hadn't realized audio data is less regular than most non-audio data files, though I can see the structured patterns in non-audio data could be more varied or complex, since they are not limited by the constraints that exist for music (rules of harmony, types of instruments, mechanics of vibration and hearing and medium). But I would have thought that music also can have "repetitive patterns spanning larger ranges than can be found by brute force search." Perhaps that's not something perceptual audio encoders can capture (but certainly evident in the sheet music of, say, a Bach cantata), since compressors are looking for simple or low kinds of patterns--can't compress by "understanding," only by math on patterns of numbers though that math might reflect or result from higher patterns.

I still don't see why it might not turn out that the types of patterns in one kind of music could in principle be more efficiently compressed using one method than those in another, even though non-audio data has as you observe more regularity. (Also, couldn't lossy encoders enhance the possibility of finding regularity, drawing them closer to non-audio data, since you have discretion to discard some of the data and smooth things out a bit.) (btw, don't know what "infinite problem space" is. and maybe you're using "complexity" in a particular technical sense.)

Public Listening Test [2010], Item Selection

Reply #51
You cannot search long ranges in polynomial time for arbitrary, repetitive patterns. But this is getting too OT. Feel free to open a separate thread.

Public Listening Test [2010], Item Selection

Reply #52
you're right it's too OT so I'll stop here.  (Looks like it might require having some detailed technical knowledge.)

Public Listening Test [2010], Item Selection

Reply #53
OK, the list of test items is almost ready. Details on each item will follow later.

The crosses indicate the type of artifacts that can predominantly occur. Note that depending on the encoder, there might be other artifacts in the decoded samples than the ones mentioned in the table.

Code: [Select]
.________________________________________________________________________
|                 |                  |                  |                |
| Test Sample     |    Transients    |   Stereo Image   |    Tonality    |
|                 | Lo-Freq  Hi-Freq | Direction  Width | Noise  Tremolo |
|_________________|__________________|__________________|________________|
|                 |                  |                  |                |
| Kalifornia      |    +        +    |                  |                |
|                 |                  |                  |                |
| Robots          |    +             |                  |           +    |
|                 |                  |                  |                |
| Applause/Harlem |             +    |              +   |                |
|                 |                  |                  |                |
| Fallen/Linchpin |             +    |              +   |                |
|                 |                  |                  |                |
| Can't Wait      |                  |              +   |                |
|                 |                  |                  |                |
| Memories        |                  |              +   |                |
|                 |                  |                  |                |
| Waiting         |                  |              +   |                |
|                 |                  |                  |                |
| Since Always    |                  |     +            |                |
|                 |                  |                  |                |
| Hancock         |                  |     +            |                |
|                 |                  |                  |                |
| Berlin Drug     |                  |                  |   +            |
|                 |                  |                  |                |
| Trumpet/Rhumba  |                  |                  |   +            |
|                 |                  |                  |                |
| Creuza          |                  |                  |           +    |
|                 |                  |                  |                |
| Girl            |                  |                  |           +    |
|                 |                  |                  |                |
| Metropolis      |             +    |                  |           +    |
|                 |                  |                  |                |
| SQAM Selection  |             +    |                  |   +       +    |
|                 |                  |                  |                |
| (Velvet,spahm)  |                  |                  |                |
|_________________|__________________|__________________|________________|


It's a total of 16 items, 15 seconds each. Makes exactly 4 minutes of material. Velvet/spahm are still open for discussion, since I don't know what to put here. It doesn't have to be one of those two, though, they are just a suggestion. Does anyone know an item not mentioned yet on which both nero and QT/iTunes do really bad at 128kb?

Edit: Let's put a deadline for the final decision on item #16 (Velvet,spahm): April 1, 2010

Chris
If I don't reply to your reply, it means I agree with you.

Public Listening Test [2010], Item Selection

Reply #54
I'm astounded what  elaborate contributions this test gets.

Public Listening Test [2010], Item Selection

Reply #55
Does anyone know an item not mentioned yet on which both nero and QT/iTunes do really bad at 128kb?


Human Disease and Techno Pop sounds pretty bad at 128kbps, on iTunes, QuickTime and Nero AAC.
"I never thought I'd see this much candy in one mission!"

Public Listening Test [2010], Item Selection

Reply #56
Sorry, I should document better my motives for the above item selection. In the course of this discussion, I was convinced that "rare extreme", "artificial" samples should be excluded since they are indeed quite far from typical musical genres such as pop, rock, and electro. Hence, I excluded items such as emese, bibilolo, lead-voice, aquatisme, all the ones in post #1 of this thread, and many of /mnt's "zappy" items (sorry, but I hope you agree that such sounds are extreme, rare, and unusually artificial). There is one exception: the two Kraftwerk items because those also test for tonality and low-frequency transient artifacts. In addition, I believe that something like "some guy playing three notes on a violin in his bathroom" is a valid, natural critical sample, so I included a few of those (note that such samples are still the minority).

I'm aware that emese, bibilolo, etc. are "classic" killer samples, so if someone wants to do a separate test on such items after this test, I'm happy to help and listen.

Conclusion: To anyone having proposals for item #16 (Velvet,spahm), please make your suggestions accordingly.

Thanks,

Chris
If I don't reply to your reply, it means I agree with you.

Public Listening Test [2010], Item Selection

Reply #57
Chris,
Can you please upload listed samples in .rar somewhere?
Thank you.

Public Listening Test [2010], Item Selection

Reply #58
Not yet, since I haven't actually constructed the samples. But I can give you the references to each item used in the samples. Sorry such details trickle in slowly, but that's all the time I have per day.

Kalifornia: first 14.6 seconds of Fatboy:
http://www.hydrogenaudio.org/forums/index....st&p=681517

Robots: cuts from beginning and mid part of the Robots remaster:
http://www.hydrogenaudio.org/forums/index....st&p=695016
Maybe this will be concatenated with a similar item if I can find one.

Applause/Harlem: concatenation of applaud00 and Harlem:
http://www.hydrogenaudio.org/forums/index....st&p=688933
http://www.hydrogenaudio.org/forums/index....st&p=621624

Fallen (should be Fall)/Linchpin: concatenation of Fall of Life and Linchpin:
http://www.hydrogenaudio.org/forums/index....st&p=195618
http://www.hydrogenaudio.org/forums/index....st&p=682220

Can't Wait: 15-second passage from Can't Wait until Tonight:
http://www.hydrogenaudio.org/forums/index....st&p=195478

Memories: 15-second passage from Cats musical:
http://www.hydrogenaudio.org/forums/index....st&p=692203

Waiting: first 15 seconds from Green Day's song:
http://www.hydrogenaudio.org/forums/index....st&p=681517

Since Always: first 15 seconds from that song:
http://www.hydrogenaudio.org/forums/index....st&p=688933

Hancock: excerpt from a Jazz piece by Herbie Hancock:
http://www.hydrogenaudio.org/forums/index....st&p=681520
Maybe this will be concatenated with a similar item (if I can find one) since it is shorter than 15 seconds.

Berlin Drug: excerpt from an unknown song (does anyone have more info on this?)
http://www.hydrogenaudio.org/forums/index....st&p=682234

Trumpet/Rumba: concatenation of trumpet and an excerpt of Entierren con Rumba:
http://www.hydrogenaudio.org/forums/index....st&p=682125
http://www.hydrogenaudio.org/forums/index....st&p=201619

Chris
If I don't reply to your reply, it means I agree with you.

Public Listening Test [2010], Item Selection

Reply #59
Continued:

Creuza: first 15 seconds of Creuza de mä:
http://www.hydrogenaudio.org/forums/index....st&p=362944

Girl: first 15 seconds of that song (Igor, what's the full name, artist, and origin?):
http://www.hydrogenaudio.org/forums/index....st&p=683378

Metropolis: first 15 seconds of Kraftwerk's song from the Man Machine, again the remaster:
was from http://www.hydrogenaudio.org/forums/index....showtopic=77128,
I don't want to re-upload it, it will become available when I present the finished test samples, anyway.

SQAM Selection: a concatenation of items from the Sound Quality Assessment Material CD released in 1988 by the European Broadcasting Union:
http://tech.ebu.ch/publications/sqamcd
This collection is widely used in the scientific literature, so why not include it here. It's now available free of charge for research and development purposes! The tracks used will be 32 (triangle), 35 (glockenspiel), and 40 (harpsichord).

--

I also found another candidate for item #16: The 15-second passage starting at 7.0 sec of Flyin' to Fly:
http://www.hydrogenaudio.org/forums/index....st&p=363343
It's quite a general-purpose critical sample, tests SBR (which we don't have here), tonality, and pre-echo issues.

For completeness:
"Roel's infamous Velvet" and spahm: http://lame.sourceforge.net/download/samples/

Chris
If I don't reply to your reply, it means I agree with you.

Public Listening Test [2010], Item Selection

Reply #60
Song Title: Girl
Artist: Beck
Album: Guero (2005)

Public Listening Test [2010], Item Selection

Reply #61
Vocal sample with very rich frequency content.
roter sand

Chris, how many samples will be included? 16-17? 16-18 is fine.

Public Listening Test [2010], Item Selection

Reply #62
I was aiming for 16, like I wrote in post #15. IMO, even that is already on the high side, and more would be too tiring for the listeners.

Chris
If I don't reply to your reply, it means I agree with you.

Public Listening Test [2010], Item Selection

Reply #63
I've tried Velvet and Fly to Fly sample. They aren't  killer samples for me.
First sample from http://www.hydrogenaudio.org/forums/index....showtopic=75868
presents problem at least for 2 AAC encoders.

Public Listening Test [2010], Item Selection

Reply #64
SQAM Selection: a concatenation of items from the Sound Quality Assessment Material CD released in 1988 by the European Broadcasting Union:
http://tech.ebu.ch/publications/sqamcd
This collection is widely used in the scientific literature, so why not include it here. It's now available free of charge for research and development purposes! The tracks used will be 32 (triangle), 35 (glockenspiel), and 40 (harpsichord).


There's a companion document that goes along with the SQAM disc. EBU truly kicks a$$ for releasing this gem to the public.

Public Listening Test [2010], Item Selection

Reply #65
So if all goes as planned when will this test begin and how long will it last?

Public Listening Test [2010], Item Selection

Reply #66
Now we're almost done with selection of samples it's time to prepare packages, web page and start test.


Public Listening Test [2010], Item Selection

Reply #68
Has this test started?

Public Listening Test [2010], Item Selection

Reply #69
Too bad that there is nobody at the moment with enough interest and free time to organize the test.
We miss you Sebastian and Roberto