What's an example of a "good DAC"?

[bigshot]

The scale doesn't include a frequency number for the point where it cuts off. Is this an old DAC without oversampling or something like that?

I'm not sure whether that would be audible or not. It's certainly marginal, but it's stone flat out to 10kHz and it looks like by the time the rolloff gets to -3dB, which is a general rule of thumb for audibility of a frequency imbalance with music, it's up to about 17kHz, which is above the range where there's much content in music. A modern DAC with oversampling really should be flat from 20 to 20, but a roll off like that is as close to the line of audibility/inaudibility as you can get. I would bet that someone could use that DAC every day to listen to music and never notice that rolloff.

 

[pinnahertz]

The person monitoring the test sets a starting position for the switch at random. You do a few passes at it. A ballpark level match is fine for the large differences he's claiming. If it's under a dB, it's fine for a casual test. These things only become sticky swamps if you go overboard on the technical details. If a casual test doesn't clearly reveal it, it probably doesn't matter anyway.

If you do the sighted test above, and recognized, in detail, what the potential biases are, it's possible to get some interesting information, but nothing you can actually put much claim in.

"Ballpark" level match is NOT fine! Less than 1dB is completely detectable. Besides, if they don't match with 1dB, what do you suggest they do about it? Test anyway?

See, this kind of thinking is every bit as bad as the "I know what I hear" stance. You're introducing clear and obvious biases...and don't care. Your results will be pretty much the result of biases.

I think a lot of the things we attribute to unconscious bias around here isn't unconscious at all. A lot of people want to game the results to feed their ego. If they want to do that badly enough, they're not going to learn anything from the test anyway. Casual tests aren't to prove things to other people. They're to find out for yourself. If you don't want to know and you just want to insist you're right without putting yourself to the test, then level matching precisely and going double blind isn't going to help. People have to start from a place of intellectual honesty.

All of the above are examples of unconscious bias, simply because the biases are not recognized as biases.

 

[bigshot]

I know what 1dB sounds like. It doesn't matter to me. I'm sure our resident 120dB friend would say that it needs to be a few digits to the right, but who cares. Detectable isn't the same as important. I don't even think 1dB is detectable in music. All my tests are with music.

 

[pinnahertz]

Yes, I know it doesn't matter to you. But if someone wants to do a real audible comparison between devices with little to no difference, like DACs, then it's really important to get a tight level match. 1dB is detectable in music, and it will skew results enough to blow up the possible 50/50 guess results. I'm not suggesting any other test signal other than music.

I'm afraid I don't understand the demand for a real blind test, then trivialize the very things such a test strives to control.

 

[castleofargh]

I often only use a switch and matched levels for personal curiosity, but I wouldn't use that as evidence to make claims. it's basically just improved sighted test ^_^.
IMO there is nothing wrong with simpler tests for unimportant stuff, but there is no working around a properly controlled test if we wish for high confidence results. people who put too much confidence in crappy tests is what got us in this mess in the first place.

 

[RollsDownWindowsManually]

I just use an A/B switch and have a 2nd person designate A or B as X for each trial. I have them flip the switch rapidly a dozen times or so between each trial, unless I was a rainman level savant I couldn't count the clicks. All conducted while I'm blindfolded of course (or at least facing another direction). I know it's not perfect science, but the best I can do. With that said, I've discovered that this 2nd person has a very limited amount of tolerance for sound science, so I can only do about one 12 trial ABX test per year, haha.

 

[Arpiben]

@pinnahertz

Ideally how will you suggest to adjust/calibrate levels ? I am assuming something like:

• digital white noise at input,
• level measurement in a 0-20kHz bandwidth
• LP analog filter for rejecting frequencies above 20kHz
• Passive attenuators for matching levels <0.2dB
• etc...

I am conscious it may be more complicated than one may think and therefore I am interested in your opinion.

 

[pinnahertz]

@pinnahertz

Ideally how will you suggest to adjust/calibrate levels ? I am assuming something like:

• digital white noise at input,
• level measurement in a 0-20kHz bandwidth
• LP analog filter for rejecting frequencies above 20kHz
• Passive attenuators for matching levels <0.2dB
• etc...

I am conscious it may be more complicated than one may think and therefore I am interested in your opinion.

That's pretty much it, though with devices like DACs that should be basically flat response, you don't really need more than a 1kHz tone. Matching levels with non-flat devices like speakers and headphones would entail more like the above, but in the end you match mid-band, where hearing is most sensitive and critical, and let the LF and HF differences just be different.

But you do need to measure, and respond to the measurement with a properly designed passive attenuator which, in some circles, will be challenged as a source of audible difference. It's tricky, and the entire circuit needs to be considered. (edit) It's sometimes necessary to put similar attenuators in the feeds to both DUTs.

 

[TYATYA]

I try to compare dac inside hdvd800, which is BB PCM1792A to the one in SP1000 : AKM4497SEQ.
SP1000 -> toslink -> hdvd800 -> hd800s (1) and SP1000 -> 2-XLR audio in -> hdvd800 -> hd800s(2).
There is no different. I can not classify them.
Other try, I replace hd800s by Shure kse1500.

I fail a gain.
Last try, I use AK120ii to feed digital to hdvd800 and use sp1000 to feed it in analog. Both dap use a same track and play same moment. Verify sq by kse1500.

I fail again.

This a 1st I found the dac in hdvd800 is at that good. Refer to its document, I saw a very good specs

 

[bigshot]

If you're trying to discern a difference between two nearly identical (or actually identical) things, and you are trying to prove something for someone else, a tight test is necessary. But you don't need a tight test like that to determine that the difference isn't worth worrying about. I test regularly for myself for spot checking stuff and comparing the effects of EQ adjustments, and that level of tightness just isn't necessary. I think it would be better if more people did more loose tests for themselves more often. It's the best way to get a feel for how numbers relate to what your ears actually hear. I've learned perspective for what a dB sounds like and how to identify imbalances in specific frequency bands by doing quick informal tests. I never would have gotten that sort of insight without my switch box.

 

[TYATYA]

My judge PCM1792A = AKM4497EQ > CS4398.
The 1st one I cant hear any difference. But 3rd one is loser in comparision.
Lastly I find those dac specs, see a similar (near the same) for both winners and they are much better than the remain one.

Simply say, dacs are same or not same sq depending on themself from model to model, the hp to verify and your ears too. Stock inear of iphoneX for exam, wont let you know the different btween dacs

 

[JaeYoon]

For me, I don't see it the same way. the Cirrus Logic CS4398. IMO I don't see the higher specs from the other two dacs are better.

From what, I see from my view is CS4398 already has specs beyond the average human hearing in average listening conditions.
DR 120, THD+N is good enough, sample rate max is 192. 24 bit. Well enough for consumer listening.
Higher specs for me don't equate to be being better. That's how companies who make newer dacs with specs on paper sell more. People are suckered so easily into buying latest 9028 pro model, or whatever is coming out with huger numbers. That's how they stay in business I guess.

 

[TYATYA]

For me, I don't see it the same way. the Cirrus Logic CS4398. IMO I don't see the higher specs from the other two dacs are better.

From what, I see from my view is CS4398 already has specs beyond the average human hearing in average listening conditions.
DR 120, THD+N is good enough, sample rate max is 192. 24 bit. Well enough for consumer listening.
Higher specs for me don't equate to be being better. That's how companies who make newer dacs with specs on paper sell more. People are suckered so easily into buying latest 9028 pro model, or whatever is coming out with huger numbers. That's how they stay in business I guess.

Not only dB is specs of a dac. There's timming in ps or ns unit.... very tiny piece of time, and 20 is not 4 even both value are so small.
120dB, like spl of a thunderbolt it mean heavy impact. Music listening usually about 110 or less.... but I feel difference impact on a better spec dac.
Think as you then all dac reach 110 will bring same sq

 

[JaeYoon]

Not only dB is specs of a dac. There's timming in ps or ns unit.... very tiny piece of time, and 20 is not 4 even both value are so small.
120dB, like spl of a thunderbolt it mean heavy impact. Music listening usually about 110 or less.... but I feel difference impact on a better spec dac.
Think as you then all dac reach 110 will bring same sq

Understandable.

Luckily for me, I'm not an audiophile and don't worry about such things.

 

[bigshot]

Not only dB is specs of a dac. There's timming in ps or ns unit.... very tiny piece of time, and 20 is not 4 even both value are so small.
120dB, like spl of a thunderbolt it mean heavy impact. Music listening usually about 110 or less.... but I feel difference impact on a better spec dac.
Think as you then all dac reach 110 will bring same sq

I think we're talking about audible differences that reveal themselves in real world situations- listening to music in the home. You're talking about the extremes of specs that would only be audible with test tones and unreasonable volume levels. Also, the peak level of all DACs is the same. The difference in signal to noise is only different at the very lowest volume level. Your thunderbolt analogy isn't correct.