using the test i did (Writing the difference between the source and encoded) you can tell how good the DualStream is. Has anyone else seen this?

What do you mean by writing the difference?

I agree though, Dualstream is impressive. B)

Ok, get the source audio file, then encode it. Find the difference between the two, and save it. (ie, if its lossless, no difference is saved, if its crap lossy, u get a lot saved to difference)

When it's good lossy you get alot saved to difference, but you don't hear this difference when the compressed file is ABXed with the original.

yeh, u kno wot i mean.

English please... This issue is discussed here before.

@den: Why don't you conduct DualStream vs. WavPack test? Now that everyone is conducting some tests these days and you seem to be the non-perceptual codec enthusiast around...

You CAN'T find/learn anything about the quality of a lossy encoded file by observing what's been thrown away during the encoding proccess. If you use the search function you will find many other discussions about this matter.

u can actually

I really would like to see a proof for this claim. (And I believe that many of the members of those forums would like the same.)

Trying to avoid a meaningless yes/no discussion, but as you can find in many earlier threads, the difference signal tells you nothing about perceived audio quality of the lossy compressed file. Must be stated here that psychoacoustic and non-psychoacoustic codecs will show different results in such a test.
Magic word here is 'masking'

[edit] This reply actually fits to the other 'difference file' you posted as well

I don't think you can infer nothing by observing the difference. Actually basing all the tests on listening could also be argued to be a flawed mentality, it's basically the opposite extreme. I think there's a lot to deduce algorithmically (like eaqual does) about how good a lossy codec is doing (not necessarily a perceptual one) by observing the encoder noise and comparing it with the original. Unfortunately AFAIK not many sound measurement methodologies/mathematical techniques regarding such comparison seem to exist. I wish all listening tests were supplemented with such additional data regarding the information theory side of compression... Such methods also have a solid advantage: It is objective and does not depend on the listeners qualities, mood, etc...

When using lossy codecs, I am not usually concerned about how a select group of bat ears in the world would do in ABX testing comparing the result vs. the original. I am more concerned about how much objective loss is incurred (because the quality at which I encode are usually beyond transparency levels). Although I know no such algorithmic measurement methods could guarantee how perfect the result would sound (perceptual side), they could still help us understand in various ways how good our approximation is. This is important especially for the reasons discussed so far like: post-processing, DSP application, transcoding,...

I think with the most files it is hard to rate their quality cause psychoacoustic is used.
Don't know what perceptual coding is (hehe, those stupid muthfckas really annoy with their insignificant posts, he?) but I guess it is nothing psychoacoustic-related.
So you are able to calculate the difference to the orig. file to measure how far you are away - I guess this difference can be ranked - maybe it has to be ranked by relativ numbers because an absolute number will be meaningless: a step away from any 10kHz tone will not sound as bad as a step away from a 4kHz tone.
Any difference can labelled as 'noise to the original' and depriciate (devaluate) the actual (means orig.) information of course.
And, what's the nutrient content of this reply - approximately zero

keep on rockin' guy

edit: maybe you have to change the numbers (4 and 10), I'm a bit clouded right now.
...just like yesterday or the day before yesterday or the day before the day bef......

EAQual results can differ from listening test results by a great deal. Please use the search function and read some of the threads.
For starters (not sure if the links will work due to server change):
http://64.246.62.80/~hydrogenaudio.org/ind...st=0&#entry5731
http://64.246.62.80/~hydrogenaudio.org/ind...st=0&#entry5918





<sarcasm>You might as well use mpc for comparing quality. Encode the original and transcode your test file to mpc. Compare bitrates. The closer the bitrates, the better. You can do some weighted delta kbps based on overall bitrate and calculate a cool score. Lower bitrates means the sample is easier to encode and the original codec has smoothed the sample. Higher bitrates mean the sample is harder to encode than the original and is thus higher quality than the original.</sarcasm>


...said Microsoft and prooved to the world that 64k wma is better than 128k mp3.

EAQual was a tool in alpha stage. And we both agree that it is a tool. But many such good algorithmic tools could be implemented that could give great insight. Because they don't depend on the mood of the listeners, have no placebo problem, ... It could make the whole process of comparison more scientific and objective. That said, I agree that such methods make sense when complemented with listening tests...

Ooh the ambiguity

(edit: haha sorry, I simply couldn't resist)

You want fries with that?

I did actually post some Wavpack vs Dualstream tests in the original Dualstream announcement thread, and concluded that for a given bitrate, Dualstream is usually more transparent than Wavpack lossy in the cases I tried. I have done some more testing since, but haven't posted my results. They have been consistent though with the above findings.

In very general terms, for my ears, Dualstream hits transparency ~50 kbits lower than Wavpack lossy for the same sample in the 275 - 400 kbit range. It also has the advantage of having quality and true VBR modes, and Florin's method of measuring the quality of the resulting file produces very consistent results in quality mode. Once you get above 400 - 450 kbits, they both get very close in transparency as you are now approaching lossless compression anyway, for some samples.

So why hasn't my signature changed? I still use Wavpack lossy because my CPU is sub 1GHz , and on my PC, Dualstream is too slow for my daily use.

Also, while I can't speak on behalf of David, Wavpack is in a state of development, with experimental noise shaping options, etc, plus a full rewrite with 4.00 on the horizon with VBR/quality options most likely, so I'm now going to wait for further development and recompare the two.



I agree with this, except that in the case of Wavpack and Dualstream, the difference or correction file can be kinda interesting. If the correction file is HUGE in relation to the encoded file, it can sometimes mean that there was a fair amount of error in the prediction that the encoder now has to hide. Sometimes with real quiet tracks, a Wavpack lossy/Dualstream encode can be nearly lossless, with only a very small correction file, and sure enough, when you listen to these, the resulting file will be transparent at a lower bit rate... B)

But what good is that, if the artificial ear hears differently than a human's ear? It's just a model, an approximation. The approximation may be good enough on a coarse scale, but when you increase "the resolution" (ie. go for near transparency, very subtle differences) the prediction error will be greater than the value you want to measure. It has been shown that EAQual doesn't offer the needed "resolution". It's like measuring the diameter of your hair with a regular ruler.