Integer is King? - 2nd EQ challenge

[Christian Budde]

Hi there,

since the last challenge was still a bit open in the end, I decided to do another challenge. It's basically double precision vs. int64 (no other algorithm included this time). However there are 16 switches (A-P) and there is at least one algorithm double and at least int64. If someone here still likes to prove if he can hear the difference, here's the chance.
The algorithms are kept nearly equal (see Integer is King? - final thoughts about the EQ challenge). I only fixed some minor bugs and I do now process both algorithms side-by-side so that it is impossible to compare it by CPU usage now. With this the switching between the algorithms is also improved. Furthermore I obfuscated the related part making it harder to identify the disassembly. In my opinion it's pretty hard to cheat now, so here's the challenge:

If anyone can really hear which one is float/integer, post it here. I am biased this time to do the test, since I know what is what, but feel free to join the challenge.

I doubt, that anyone will be 100% right in this test, but even 90% would be pretty good.

Here's the TestEQ: http://www.aixcoustic.com/TestEQ2.zip

Good luck,

Christian

PS: If you're only sure about some algorithms, because the listening tests are too exhautive, feel free to only post the preliminary results.

[thorkz]

Thank you much Christian. Very appreciated.
Thats a tough one, but lets try.
So only int64 and floatdouble.

thorK


By the way, if anybody hasnt yet, check out the two output files from acidose' filter in the other threat.

[defjamm]

By the way, if anybody hasnt yet, check out the two output files from acidose' filter in the other threat.

which are flawed. read what JonHodgson wrote.

[thorkz]

Well. I have already proven to have some difficulties with int64 and float. Right now it seems to me the future development of audio soft into higher and higher bits is not what will make me happy all around. I still find your 32int filter very nice and musically and much easier to tell apart from the others.

So far I have a few favorits, but I have to change to headphones cause my speakers dont cut it anymore for this. Headphones give me eartones though, so that it might still take a little to come to my final result. Its seems to become more an obversation of what the sound does to me than an observation of the sound itself.

But the diffs are there.

Here my farorits: ( THIS IS NOT MY OFFICIAL STATEMENT )


The first ones I recognized as int and that seem to come back are B and O.
N seems to be in there and has a fantastic high end.
About F i am still very unsure.

J seems to change . (you wouldnt do something this nasty would you? yeah you would )


So far. We'll see what the headphones will do.

[END OF UNOFFICIAL STATEMENT]



thorK

[thorkz]

I think I'll have to keep you [UNOFFICALLY]updated since headphones dont seem to improve the listening situation very much. In fact I cant get that bass feel into my stomach with headphones so they all sound quite nice except for A,C,D and P which are most likly floats. There seem to be other floats involved too, that sound quite nice on headphones but cant be int because I didnt recognize them as ints on the speakers yesterday, if that makes any sense. I'm still learning what to listen for with these int64s.

Its not much fun, it's hard work.


So far I think: B, F, O are int. Remaining canditates are: N and ? what else. I have to put this all into some structure.

[End of not existing post]

thorK

[thorkz]

I give up. I cant find any more ints plus now I am really starting to hear unreal diffs that are there and then they are not but there they are again. I am just surfing on the changes from filter to filter and cant concentrate on what I want to check. I guess thats what double blind is all about?


So here is my result ( mauseleum please correct me if im wrong!)

Ints: B,F,O i found N and K to sound really interesting so I take them in there.

The others are floats then
which are: A,C,D,E,G,H,I,J,L,M and P


Soo. What do you say Mauseleum? If you done the test please tell us what you heard. I'm through with this. My nervs tell me to.


thorK

[bmanic]

Yay! More challenges. Downloading EQ now. This time I'll take it to the studio and listen on the genelecs!

Cheers!
bManic

[thorkz]

So.Christian? There seems not to be too much interest anymore?
Can you unveil the solution?

thorK

[druid]

I'd love to do this if I had time, and if I could hear a difference. But I assure you I would not; my ears aren't trained well, and my equipment is sub-par in almost every respect, if I wanted to do this.

I wish a couple more would chime in so that I can see how many people believe there is no difference between that many different kinds! I'd like to see a person who says they can hear a difference get it completely wrong. :D

[Lance]

I pass the task to batman, he has better ears.

[sambean]

My pet cat reckons there's 2 with integer, one low option, one high option..

[Christian Budde]

I will unveil the results tomorrow. However if there's someone who likes to take part, now is the chance!
Christian

[Lance]

I've tried it. My conclusion is that what I expected; no difference, I can't hear difference, though I have good ears and I used a good pair of AKG headphones. Maybe batman will.

[Christian Budde]

Ok, here are the results: A & K are int64, the rest are doubles. If someone doubt it, just gain the signal prior to the input by 48dB or more and attenuate it after that. Both integer version will result in heavy clipping, while the float version will remain uneffected.
From this I came to the conclusion, that even thorKz can't hear the difference between the format itself. All further difference must be due to implementation difference (bugs?) between an integer and float versions (see the other threat).

For me, the file is closed now.

[advaya]

I pass the task to batman, he has better ears.

IMAGE REMOVED FROM QUOTE (BECAUSE IT'S JUST TOO PAINFUL TO SEE TWICE IN ONE THREAD)

I think his super-hearing is the result of his super-bat-costume-nipples.