[sandslash]

Whereas 16-bit is enough for the playback format, I argue that if one uses digital volume control it is still better to use a 24-bit DAC. Every time you halve the digital volume control, you truncate one bit of dynamic range. Depending on the peak listening volume of yours, if you have the downstream amplification at high gain and digital volume control really low before it, the quantization error can become an audible problem at some point if you use a 16-bit DAC.

 

[xnor]

Quantization distortion is eliminated using dither. But I agree, high gain after lots of digital attenuation will cause the noise floor to become audible.

 

[Speedskater]

In another thread, in another forum, on another day they discussed volume controls.  It seems that many components volume controls aren't what they seem. Some are true digital, some are digital controlled analog, a few are analog and some are combinations.
The problem gets deeper when one control is used to adjust both stage gain and volume.

 

[spagetka]

I am sorry but I do not understand what is going on here. Maybe I am still missing something here.
 
I chose the track and you filters and prepared the files. As you said before, filters should not make the difference because sample does not contain much information above 10kHz.
 
First, try to find 2 people with the score 7 and more out of 10 and then we can talk about (non) significant result. Or even try to pass the test with random selects and hope that you will be lucky.
Second, of course I found artifacts (do not know if it is the right word but for me it is "something" different) otherwise I think that I cannot pass the ABX test.
Third, I cannot identify which track is currently playing, I can say that there is a difference between them. 
Fourth, and for me the most important, we are discussing the difference between the files based on listening not trial/error method.
 
Sorry for my bad English, have a good day.
 
Quote:

 
That is fine, but cherry picked results are not suitable for calculating an overall score (like combining the two "good" tests into 15/20).
 
 
7/10 is surely not enough, and 8/10 is still above the p-value of 0.05 which is the "standard" threshold for a statistically significant result. Given that for the "7/10" test you also had an aborted 4/8 score, I do not think that one is good enough. 8/10 is marginal, and needs more testing for a more definite result either way.
 
Did you find some artifact in any of the files that you specifically listen for when performing the test ?

 

[chewy4]

Quote:

 
Or even try to pass the test with random selects and hope that you will be lucky.
 

Ok, I just did.
 
 
foo_abx 1.3.4 report
foobar2000 v1.1.14a
2013/03/10 13:29:48
 
File A: C:\Users\Computer\Music\[demonoid[Mount Kimbie - Crooks & Lovers (2010) [320]\01 Tunnelvision.mp3
File B: C:\Users\Computer\Music\[demonoid[Mount Kimbie - Crooks & Lovers (2010) [320]\02 Would Know.mp3
 
13:29:48 : Test started.
13:29:49 : 01/01  50.0%
13:29:50 : 02/02  25.0%
13:29:52 : 03/03  12.5%
13:29:53 : 03/04  31.3%
13:29:55 : 04/05  18.8%
13:29:56 : 05/06  10.9%
13:29:58 : 06/07  6.3%
13:29:59 : 07/08  3.5%
13:30:00 : 08/09  2.0%
13:30:02 : 08/10  5.5%
13:30:05 : Test finished.
 
 ---------- 
Total: 8/10 (5.5%)

 
I didn't listen to these files. I just selected the top option every time. Got 7 out of 10 on my third attempt. 8/10 on my fifth or sixth.
 
If you have to cherry pick your results, it generally means you're not doing so well. You need to display a level of consistency to show that you are hearing a difference.

 

[stv014]

Indeed, those not familiar with ABX tests (or statistics) often think that a simple "majority" result is good enough, but with a small sample size that is too easy to achieve by random guessing.

 

[mikeaj]

Yeah, there's some serious data snooping if you're just picking out favorable results.
 
Getting to 5% on the ABX test means that somebody picking by complete chance (flip a coin, or whatever) would achieve the same score as you just did or better given that many trials, 5% of the time.  It's not so strong a statement.  And if you have to pick out of a subset of results, that's much weaker statistically.

 

[xnor]

Quote:

Third, I cannot identify which track is currently playing, I can say that there is a difference between them.

This doesn't make sense. Either you can identify X as either A or B because you can hear a difference between A and B or you cannot identify X because you cannot hear a difference between A and B.

 

[scuttle]

Quote:

Look, it is only about the concentration. Of course I had unsuccessful attempts before because I am not the robot and which is more important I try to find the weak point in random part of the test track and of course it takes a while to have ears adapted.

 

 
If you keep on guessing at random until you get a set of answers right by sheer chance, and count only that last set, then it is impossible to fail! What you did was meaningless. If you wanted practice at training you ear before taking the test then you should have used a different track and then ***decided in advance that the first attempt made with the new track would be the one that counts.***

 

[spagetka]

)))))))))))))))

 

[leogodoy]

Keep ABXing like that and you may be THE ONE who will "prove" paying large sums of money for headphone or USB cables do make a difference.

 

[dynamics]

Are there any cons to just leaving the DAC at 24bit 192hz?  Or would 16bit 96hz be ok?  I really can't hear the difference by the way.  My hearing is good too.  But I just couldn't distinguish one from the other.

 

[jcx]

24 96 is often considered the sweet spot - modern flagship DAC do deliver effective 18-20 bits of S/N performance which can be used by modern sw digital volume or other audio processing - but dithered 16 bit will usually be OK with even half right system gain structure
 
192k for some DACs shows worse S/N and distortion than at 96k - there will always be some speed penalty in digital noise, power requirements that can impact analog output

 

[harmonix]

Quote:

Quantization distortion is eliminated using dither. But I agree, high gain after lots of digital attenuation will cause the noise floor to become audible.

 
An ignorant question - dither/white noise is added which is supposedly uncorrelated with the signal.
Well we all know these cannot be truly random sequences. So how random is this white noise and how do we really know it's uncorrelated? 
 
Sorry if this has been asked more than once.

 

[stv014]

Quote:

An ignorant question - dither/white noise is added which is supposedly uncorrelated with the signal.
Well we all know these cannot be truly random sequences. So how random is this white noise and how do we really know it's uncorrelated?

 
It can easily be made random enough that it is not practically different from "ideal" white noise for the purpose of dithering, unless the input signal already contains the same pseudo-random sequence for some reason. That has virtually zero chance of happening accidentally, so it is normally only an issue if the signal was already dithered once with the same noise, or possibly in the case of noise shaping, where the dither is in a feedback loop. For simple dithering in software, the pseudo-random generator can be initialized from the current system time to avoid the problem of dithering more than once with exactly the same noise.
 
As an example, in the audio processing utilities linked in my signature, I use this simple algorithm to generate noise:

 x[n] = (x[n - 1] * 742938285) % 2147483647

where x[n] is the current state of the generator (an integer in the range 1 to 2147483646), and % is the modulo (remainder of division) operator. The output of this passes basic tests of randomness like the DIEHARD battery of tests, and is plenty good enough for generating white noise and dithering in particular (but not for more demanding scientific or cryptographic applications). The sequence does loop after 2147483646 samples, but that is not a problem for audio (it is more than 1.5 hours of stereo noise at 192000 Hz sample rate). For the purpose of dithering, the distribution of the noise is made triangular by subtracting the previous sample, which also shifts some of the noise energy into the higher frequency range and slightly reduces the weighted level of the noise at sample rates above 32000 Hz. By default, the initial 'x' in the PRNG is set from the current system time, so the output file is always slightly different. Using some optimization tricks to avoid the expensive % operator, the dithering costs about 12.5 CPU cycles per mono sample of output.