[stv014]

Quote:

Originally Posted by oyster /img/forum/go_quote.gif
 
After listening for good 25-30 min here is my opinion (Hope C is not some portable player

 ) -

 

Spoiler: Warning%3A%20Spoiler!

I do not know what C is, but I am fairly sure it is not the original sample.

 

[jake120]

Quote:

 

Spoiler: Warning%3A%20Spoiler!

I do not know what C is, but I am fairly sure it is not the original sample.

I agree.

 

[stv014]

Quote:

Suggestion to Skamp: Have you considered picking say two players A and B and repeating the play record cycle several times on each? I.e. take your Clip, play & record - and use the result as the input to the next cycle WITH THE SAME PLAYER, then repeat several times? This could be valuable in two ways:

 
Try these difference files, they basically amplify the non-linear flaws of the system by 100 or 1000 times, while at the same time partly canceling out the original signal. This makes distortion and noise much easier to hear. Also, frequency response can easily be extracted and applied any number of times with convolution.
 
Technically, all errors are additive or multiplicative, "bottlenecking" occurs only when one error is so large that it makes the effects of other ones negligible (e.g. if the sound is downsampled to 22050 Hz, then frequency response above 11025 Hz will not matter, because that range has already been removed; also, if at one stage -60 dB noise is added, then the further addition of -100 dB noise does not matter, because it has 10000 times lower power, and will only make the SNR worse by 0.00043 dB; clipping falls into this category, too, since it removes information depending on the signal level). Some errors are reversible and in lucky cases can be partly canceled out (some might call that "synergy").

 

[miceblue]

Quote:

Quote:

Originally Posted by oyster /img/forum/go_quote.gif
 
After listening for good 25-30 min here is my opinion (Hope C is not some portable player

 ) -

 

Spoiler: Warning%3A%20Spoiler!

I do not know what C is, but I am fairly sure it is not the original sample.

To me, C seems to be the most dynamic of the 6. Whether or not that actually means anything is up in the air to me, but compared to A and G it definitely seems more natural, airy, spaced out, I can't really describe it. XD
In G, the guy's voice seems more nasally, for example, whereas in C he sounds more realistic to me (his voice has more weight or something).
 
OK since I said I would try ABX tests in a previous post:
 

Spoiler: Foobar%20ABX%20C%20vs%20G%20Test%20Results

 
foo_abx 1.3.4 report
foobar2000 v1.1.18
2013/02/04 01:44:35
 
File A: C:\Users\Michael\Desktop\C.flac
File B: C:\Users\Michael\Desktop\G.flac
 
01:44:35 : Test started.
01:47:23 : 00/01  100.0%
01:47:33 : 01/02  75.0%
01:47:40 : 02/03  50.0%
01:48:11 : 03/04  31.3%
01:48:27 : 04/05  18.8%
01:49:05 : 05/06  10.9%
01:49:19 : 05/07  22.7%
01:49:56 : 05/08  36.3%
01:50:24 : 06/09  25.4%
01:50:46 : 07/10  17.2%
01:51:03 : 08/11  11.3%
01:51:11 : 09/12  7.3%
01:51:39 : 10/13  4.6%
01:51:54 : 11/14  2.9%
01:52:11 : 12/15  1.8%
01:52:40 : 13/16  1.1%
01:52:47 : Test finished.
 
 ---------- 
Total: 13/16 (1.1%)
 

I might try A and C next.

 

[oyster]

Quote:

 

Spoiler: Warning%3A%20Spoiler!

I do not know what C is, but I am fairly sure it is not the original sample.

Yeah, C does have some added bass compared to the other ones i liked (D,A) but....hmmmm....i liked it...

 

[scuttle]

Quote:

scuttle: thanks for the suggestion, but I really can't be bothered to do that

It would be somewhat interesting, but it wouldn't help with the purpose of this test.

 
I'll probably write a script to do it automatically. After doing that it shouldn't be too hard to come up with a way of comparing the results automatically that will give a fairly meaningful single number - eg the average level of the diff file after 10 cycles. Hopefully I'll find that the result is one that you can predict by directly extrapolating from a dif between first generation files.

 

[skamp]

You're using devices not everyone is familiar with.

The people who made claims that prompted me to come up with the test, claimed to be familiar with two specific devices that are included in my test. They weren't vague about it.

Whether or not one needs to be familiar with the gear is also debatable: if the differences are as obvious as claimed, surely they would show immediately, even in a 30s sample (which comes from an excellent master, btw, and includes several instruments as well as singing).

 

[Takeanidea]

The people who made claims that prompted me to come up with the test, claimed to be familiar with two specific devices that are included in my test. They weren't vague about it.

Whether or not one needs to be familiar with the gear is also debatable: if the differences are as obvious as claimed, surely they would show immediately, even in a 30s sample (which comes from an excellent master, btw, and includes several instruments as well as singing).

 
From my point of view the lack of knowledge of the track and the sources and a certain fear of not being able to tell them apart resulted in a very negative experience where I ended questioning even my own existence in the end

 

[scuttle]

Quote:

The people who made claims that prompted me to come up with the test, claimed to be familiar with two specific devices that are included in my test. They weren't vague about it.

Whether or not one needs to be familiar with the gear is also debatable: if the differences are as obvious as claimed, surely they would show immediately, even in a 30s sample (which comes from an excellent master, btw, and includes several instruments as well as singing).

 
I think criticism of the devices chosen as "unfamiliar" is very unfair. The Clip and Pod are about the most familiar devices that could be possibly chosen- the clip is cheap enough so that most people can just buy one if they are curious - and the O2/ODAC combination will almost certainly be accepted by everyone whose opinion is worth a damn as providing a reference level. It's fascinating and worthwhile just to find out if people  can tell the difference between the Clip (the world's best cheap hw) and the ODAC/02 (a reasonable approach to perfection.)

 

[Chucky-X]

Hi Skamp,
 
I think the reason that your files sound relatively similar and do not really represent the analog output of the tested devices is because ->
 
Please take a look at this FFT from a DAC that I made in the late 1990's. This was measured on an Audio Precision System Two by Helmut van Aaken, technician at Thum + Mahr, a german company that sells broadcast equipment.
 
Here you see the harmonic structure of the overtones that are generated during reproduction. Note that those overtones are well below the quantization of a 16 bit recording.
 
They are very silent and audio engineers often like to make those harmonics as small as possible. However, the harmonic structure is where the business takes place, not only in their magnitude but moreso in their distribution as such.
 

 
On the AP shot you see the 2nd harmonic down at about -105 dB, the third at -120dB, the 4th up a little, the 5th down a bit and so on. Every odd order harmonic in my design is lower than the even order harmonic. This is what makes a sound. Or in other words, this an indication of what makes a sound good.
 
The human ear is extremely sensitive to the harmonic content of a sound, and by extremely I mean extremely !
 
Odd harmonics are perceived much louder compared to even harmonics. Odd harmonics cause alert. It is the trick of nature to percieve the cry of a baby or a small child as extremely loud and alerting although the output as such is relatively low.
 
Whenever an instrument, a voice or a DAC (as in your case) is recorded, the harmonic content falls below the quantisation, that means it is NOT on the recording. Therefore your recordings do not show much the audio quality of the DACs. At least they don't show everything there is.
 
It is the (undiscovered by most) duty of the reproduction equipment to re-create the harmonic structure, in order to achieve a good sound quality. 
 
The measurement above is a hint only, because it is primitive. It only measures the harmonic content of one steady frequency and only up to 22kHz. It does not take into account the dynamic behaviour of the harmonics with changing amplitude and frequency. And it also does not show what's happening in the MHz region as every noise is energy, and energy does not get lost into nothing.
 
This would be subject to a whole new universe of research.
 
Are you up to it ?
 
Charles

 

[stv014]

Quote:

Originally Posted by Chucky-X /img/forum/go_quote.gif
 
Whenever an instrument, a voice or a DAC (as in your case) is recorded, the harmonic content falls below the quantisation, that means it is NOT on the recording. Therefore your recordings do not show much the audio quality of the DACs. At least they don't show everything there is.

 
If the quantization noise was high enough to obscure audible distortion, then the quantization noise would be audible instead, and therefore the recorded file would still sound different from the original.
 
Quote:

Originally Posted by Chucky-X /img/forum/go_quote.gif
 
The human ear is extremely sensitive to the harmonic content of a sound, and by extremely I mean extremely !

 
That is great, it should be easy to hear the difference then.

There is plenty of measurable harmonic content captured by the recording.

 

[skamp]

Note that those overtones are well below the quantization of a 16 bit recording. […] The human ear is extremely sensitive to the harmonic content of a sound, and by extremely I mean extremely !

If it falls below 24 bit quantization (the recordings where made at 24 bit), then there's no way it's audible to begin with.

 

[Chucky-X]

Quote:

If it falls below 24 bit quantization (the recordings where made at 24 bit), then there's no way it's audible to begin with.

Then that's the reason they all sounded good on the Tera

 

[stv014]

Quote:

If it falls below 24 bit quantization (the recordings where made at 24 bit), then there's no way it's audible to begin with.

 
The uploaded files (including the source) are all dithered to 16 bits, though, even if it is unlikely to make a difference, since the sample is never quiet enough for 16-bit quantization noise to become audible.

 

[skamp]

I don't know where this is supposed to be going, since no-one is claiming that 16 bit quantization noise is audible.