Head-Fi.org › Forums › Equipment Forums › Sound Science › DAC blind test and difference extraction
New Posts  All Forums:Forum Nav:

DAC blind test and difference extraction

post #1 of 15
Thread Starter 

Here is a small set of files that tests the quality of the audio recorded from two DACs. It includes level matched and synchronized samples for ABX testing (I do not think there will be many positive results, though, maybe some for the onboard), but perhaps more interesting are the "difference" files, which were created by matching the levels, frequency and phase response, speed, and delay of the files as accurately as possible, and then subtracting the original from the recorded sound. This makes the artifacts of the DAC much easier to hear. However, the synchronization is not perfect (it cannot really be, since the DAC and ADC clock frequency slowly changes over time by a small amount), so it is still possible to hear the original audio, but it is significantly attenuated.

 

For more information, see also this and this post. For the Xonar D1 difference signal, I used somewhat better synchronization, but it is limited by the very low frequency DAC/ADC jitter (note how the sound fades in and then out over time as the delay between the samples changes).

 

Xonar D1

 

D1_A.flac - the original file resampled to 96 kHz

D1_B.flac - the recorded audio (level matched, pitch corrected by a factor of 1.00000592658676, and synchronized with the original)

D1_diff.flac - difference signal amplified by 60 dB

 

Realtek ALC887 onboard audio

 

loopback.7z - includes A.flac and B.flac, and various source files

diff40dB.flac - difference signal amplified by 40 dB


Edited by stv014 - 1/30/13 at 12:14pm
post #2 of 15

http://www.libinst.com/Audio%20DiffMaker.htm ?

post #3 of 15
Thread Starter 

As noted in the posts linked with more detailed information, I already tried it, but it gave worse results than the "manual" method. It was not able to detect the sample rate difference as accurately. However, I only used the program with the default settings, experimenting with the advanced options could very well have improved the accuracy.

post #4 of 15
Quote:
Originally Posted by stv014 View Post

more interesting are the "difference" files, which were created by matching the levels, frequency and phase response, speed, and delay of the files as accurately as possible, and then subtracting the original from the recorded sound. This makes the artifacts of the DAC much easier to hear.

That doesn't say anything about audibility when playing music through various DACs, though.
post #5 of 15
Thread Starter 
Quote:
Originally Posted by skamp View Post

That doesn't say anything about audibility when playing music through various DACs, though.

 

What it does show is that even if the error is amplified by a factor of 1000, and at the same time the source signal is attenuated, there are still no major artifacts revealed. It is a popular belief that cheap gear (particularly amplifiers that use strong negative feedback, op amps, and delta-sigma DACs) measures well with sine waves, but its performance degrades drastically with a complex real world signal like music, and becomes worse than that of more expensive and worse measuring "audiophile" devices (NOS DACs, tube amplifiers, single ended class A amplifiers with no feedback, etc.). For those who believe in this, and have objections to ABX testing, such difference files that "magnify" the flaws of a device could be interesting to listen to. It also makes real, existing artifacts easier to hear and analyze.

post #6 of 15
Quote:
Originally Posted by skamp View Post


That doesn't say anything about audibility when playing music through various DACs, though.

ABX it through f2k and see what you think.  I scored 3/9 - 91.1% probability that I was guessing.  Great to get some more input from others to see how audible DAC differences are to your ears.  Just in case you're not sure about ABX'ing, check out the foobar plugin at: http://www.foobar2000.org/components/view/foo_abx

 

Select the two A or B files from the above and then right click the two files and go to utilities / ABX, and it will run a ABX test which is a blind test between two samples against two other samples, randomly, with a score and probability on your guessing.

post #7 of 15

Tried an ABX - failed miserably !

post #8 of 15

Tried an ABX, got 13/15 - are you SURE these were sufficiently matched in volume?  I don't have the tools or know-how to check, but one seemed a bit louder at roughly the same frequency that bothers me about the HE90s (the one at which my ear canal resonates, giving me a bit of annoying pain)

 

Indeed, I performed this test not by truly hearing differences in the music, but by deciding which one had high piano notes that were pissing me off the least


Edited by El_Doug - 2/16/13 at 4:36pm
post #9 of 15
Thread Starter 

Did you compare D1_A.flac and D1_B.flac ? The files are level matched (ReplayGain shows a difference of 0.01 dB, and that is probably because of the small FR differences: -0.1 dB at 20 Hz, and a few hundredths of a dB in the treble). The graph below shows the level differences in D1_B.flac relative to D1_A.flac, it was created directly from those files, and is not a separate frequency response measurement:


Edited by stv014 - 2/16/13 at 5:06pm
post #10 of 15

Yeah, got the two files from your dropbox... that's really weird, maybe I just got 1% lucky? 

post #11 of 15
Quote:
Originally Posted by nick_charles View Post

Tried an ABX - failed miserably !


That's because you didn't practice.  You went into the test listening for those special "catch phrase" sounds the equipment reviewers are always writing about.  So practice by making "B" about 2 dB's louder than "A".  Can you pass the test?  How about 1 dB?  Or maybe EQ "B" and try that.

post #12 of 15
Thread Starter 
Quote:
Originally Posted by El_Doug View Post

Yeah, got the two files from your dropbox... that's really weird, maybe I just got 1% lucky? 

 

If you really hear a difference, it would be interesting to find out why. Could it be the DAC filter ? I will create another file by upsampling the original sample using the impulse response of the loopback as the lowpass filter.

post #13 of 15
Quote:
Originally Posted by Speedskater View Post


That's because you didn't practice.  You went into the test listening for those special "catch phrase" sounds the equipment reviewers are always writing about.  So practice by making "B" about 2 dB's louder than "A".  Can you pass the test?  How about 1 dB?  Or maybe EQ "B" and try that.

 

I did practice to a degree.  I repeated small sections several times directly comparing A/X. B/X, A/Y and B/Y. I do not as a rule read equipment reviews except for the measurement parts or for the purpose of having a good laugh. I suspect my failure is more related to my 54 year old ears and the magnitude of the difference(small), In the past I've correctly DBT'd CD players with a 0.7db level difference and even high-res vs redbook , but only when there was an obvious artifact such as a blip or clipping or the samples were not perfectly aligned. 

 

That El_Doug managed to tell the two apart is impressive but he hints it may possibly be due to some kind of an artifact. Back in 2000 when the first semi-serious attempts to test codecs were developed there was a German mag that did some WAV vs Mp3 tests , one person managed close to perfect discrimination, it turns out this subject had a moderate hearing loss at some frequencies, it was suggested that this "filter" allowed him to hear the artifacts typical of earlier codecs rather more easily as they were not masked 


Edited by nick_charles - 2/17/13 at 8:19am
post #14 of 15
Thread Starter 
Quote:
Originally Posted by El_Doug View Post

Yeah, got the two files from your dropbox... that's really weird, maybe I just got 1% lucky? 

 

Try comparing this new file to D1_A.flac and D1_B.flac:

 

D1_C.flac

 

It was created by resampling and convolving the source file in software so that it simulates the frequency response and DAC filter of D1_B.flac, but does not have any analog distortion, jitter, etc. It can also be subtracted from D1_B.flac, to get a difference signal similar to the one in the first post.


Edited by stv014 - 2/25/13 at 11:23am
post #15 of 15
Thread Starter 

I do not know where this should be posted, since the original thread is locked, but, since the the deadline for "Blind test: 6 DACs compared" was yesterday, I show what I guessed using measurements, in case anyone who took that test is interested. Of course, I am not sure if these are actually right, but the majority of them is likely to be correct:

F: ALC663
B: Clip+
G: E-Mu 0204
E: Galaxy Nexus
C: iPod Classic
A: ODAC
D: Original
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: Sound Science
Head-Fi.org › Forums › Equipment Forums › Sound Science › DAC blind test and difference extraction