DAC audible differences are finally proven with ABX?

[pma]

Well, mystery solved then. I would not expect any timbral shift only from the TD+N difference.

Though the TD+N difference is huge. OK, below is the link to the files with correct (same) polarity. If you wish, please try foobar abx and post the result. I will do it as well.

https://pmacura.cz/multi_corrpol.zip

 

[pma]

I will do it as well.

Quickly got this. I also have a feeling of a bit timbre difference.

foo_abx 2.1 report
foobar2000 v2.0
2026-08-05 08:17:04

File A: MT_orig_96k_PCM24.wav
SHA1: 4c4f8dec5424ed619a31c41ed27c557900d26a79
File B: multitone_bt_corrpol.wav
SHA1: 917944ba19692c365d73b8003fc745a4637bf058

Output:
ASIO : Focusrite USB ASIO
Crossfading: NO

08:17:04 : Test started.
08:18:05 : 01/01
08:18:17 : 01/02
08:18:40 : 02/03
08:18:48 : 03/04
08:18:59 : 04/05
08:19:09 : 04/06
08:19:24 : 04/07
08:19:32 : 05/08
08:19:42 : 06/09
08:19:56 : 07/10
08:20:06 : 08/11
08:20:15 : 09/12
08:20:25 : 10/13
08:20:37 : 10/14
08:20:44 : 11/15
08:20:52 : 11/16
08:20:52 : Test finished.

 ----------
Total: 11/16
p-value: 0.1051 (10.51%)

 -- signature --
743c8dc42e862bb5bb1c6bcb91fd6aef8d1ce019

 

[Thomas_A]

Though the TD+N difference is huge. OK, below is the link to the files with correct (same) polarity. If you wish, please try foobar abx and post the result. I will do it as well.

https://pmacura.cz/multi_corrpol.zip

Can look at it later tonight, but I don't expect much...

 

[pma]

BTW, there was a discussion about polarity check from the multitone record. DeltaWave can do it, it tells it correctly in its record. The files that I posted can be checked with DeltaWave.

This is how it told with the files posted yesterday (polarity inverted)

DeltaWave v2.0.13, 2026-08-04T16:52:14.4091845+02:00
Reference: MT_orig_96k_PCM24.wav[L] 1000000 samples 96000Hz 24bits, mono, MD5=00
Comparison: multitone_bt.wav[L] 1000000 samples 96000Hz 24bits, mono, MD5=00
Settings:
Gain:True, Remove DC:True
Non-linear Gain EQ:False Non-linear Phase EQ: False
EQ FFT Size:65536, EQ Frequency Cut: 0Hz - 0Hz, EQ Threshold: -160dB
Correct Non-linearity: False
Correct Drift:True, Precision:30, Subsample Align:True
Non-Linear drift Correction:True
Upsample:False, Window:Hann
Spectrum Window:Kaiser10, Spectrum Size:262144
Spectrogram Window:Hann, Spectrogram Size:4096, Spectrogram Steps:2048
Filter Type:FIR, window:Hann, taps:8192, minimum phase=False
Dither:False bits=0
Trim Silence:True
Enable Simple Waveform Measurement: False
Discarding Reference: Start=0s, End=0s
Discarding Comparison: Start=0s, End=0s
Initial peak values Reference: -2.647dB Comparison: -2.227dB
Initial RMS values Reference: -15.043dB Comparison: -15.046dB
Null Depth=12.397dB
Trimming 222 samples at start and 50 samples at the end that are below -90.31dB level
Phase inverted

 

[solderdude]

So a retest is in order with the polarity of one of the files flipped ?

 

[pma]

So a retest is in order with the polarity of one of the files flipped ?

Yes. As the BerryBak Bluetooth amp inverts polarity (verified), I have just inverted the original yesterday record from that amp in Adobe Audition and posted as a new test today, with test files in same polarity. It is an intentional part of our efforts to find possible variables that might affect the tests. We shall see if inverted polarity might have affected the result of my test or not.

 

[AndyBell]

We are trying to do that but with much difficulty due to unique nature of how the test was done with an out of production comparator

I know and it makes for sad reading - Josh could make your life so much easier if he wanted to.

 

[SSS]

If I may, your previous attempt wasn't a test finding, unless you were testing sighted listening.
Your second test with Foobar is a test finding of the files.
It is entirely possible, even likely that you thought you heard a clear difference where there was no clear difference.

Distortion artifacts are surprisingly hard to hear. Which is the point of the test. Which is why it is so darn important to do these studies with proper controls, including a proctor to ensure that the controls are adequately preventing the test from being false.

I like to think of myself as pretty experienced. But I am not able to overcome sighted bias, it rules us.

First test with the mediaplayer on Windows was not correct since it seems that it alters the sound. Foobar 2000 test shows that I cannot hear a difference between the files regardless how many times I switched. Therefore no need for a controlled blind test. If somebody younger with very good ears may hear a difference at first try then a controlled test may applicable.

 

[Thomas_A]

Yes. As the BerryBak Bluetooth amp inverts polarity (verified), I have just inverted the original yesterday record from that amp in Adobe Audition and posted as a new test today, with test files in same polarity. It is an intentional part of our efforts to find possible variables that might affect the tests. We shall see if inverted polarity might have affected the result of my test or not.

I wll also revert the original as a positive control. Tonight. (I would also say it is already concluded that some people are sensitive to polarity switches but some not. That was why I asked the question early in the thread. While Josh says no DAC inverts polarity on its RCA, there is no conclusive data shown. And no reference to any review. The D90 had RCA ans XLR swapped, and this was accidentaly found by users that used both outputs to mains and subwoofers. Hopefully Amirm will find the specific Matrix DAC and test both outputs.)

 

[pma]

It was a good question. And all the files are available - original multitone, BT record with inverted polarity and BT record with correct polarity. If we are getting statistically valid positive results, we may think of 3 reasons:

1. Inverted polarity
2. TD+N ditortion (Of the amp and Bluetooth transfer with Apple AAC chodec)
3. High frequency decay above 10kHz in the files recorded via BT (-1dB/15kHz)

We have not tested (3) yet, but it can be easily done by using LPF 10kHz filter in AA on both original and recorded files.

 

[SSS]

Listened to the recent polarity corrected tracks with Foobar and STAX electrostatic headphones. Sorry to say, I can't hear a difference even when played loud and my ears fatigued. Listened especially for low mid and high tones seperately, no way. I assume that the masking effect at least for my old ears hides this distortion mud.

 

[pma]

Thank you for taking the test. It has been my intention to show that importance of non-linear distortion is exaggerated. And we have very very high level of TD+N, about -34dB. Not -120dB, not -100dB, not -60dB.

 

[hardisj]

I have this on my to-do list. Currently I’m steeped in clothes shopping with my teenager for school. Hopefully tomorrow I’ll be able to finish up using the ABX for the current review I’m working on. Photo below.

Once that is finished I’ll move the ABX to my office to get some measurements. I will be in North Carolina next weekend for the Audio Advice live show.I’m shooting to have the ABX data posted before I leave. But I can’t guarantee it between my day job, family and other personal interests.

TL;DR
I’ll provide the data. The group here just needs to be patient.


EDIT: But let’s not forget. It’s possible - not probable - that Josh’s unit could somehow measure differently than mine. So there is still the potential that any issues with his testing wouldn’t be shown by measurements of mine. I’d be extremely surprised if it were, though.

View attachment 548674

Update:

The measurement of my ABX isn’t happening this week. I’m terribly far behind. I leave tomorrow for Audio Advice Live. And I’ve been busting my a$$ with my review of this Orsefon (Mcintosh clone) amp and I’m so far behind that I feel bad for the owner who loaned it to me and is waiting for it to go back to him. I actually shot the video last night only to then corrupt the SD card because I cut power to the camera before ejecting the card (insert obligatory “pull out” joke here). So I have to remake the damn video. Ugh.

FWIW, in terms of time, gathering and posting data to my website is kindergarten level compared to making a video discussing the product, measurements, listening, etc. It’s literally a second full time job. I enjoy it most of the time but I hate holding onto loaned gear for too long and it gets stressful at times.

But I promise the measurements of my ABX is coming eventually. I’ve just got other obligations to wrap up first.

I’m as interested in the results as anyone here. Just bear with me. Thanks for your patience.

- Erin

And no, I’m not at home where I could be working on my ABX box measurements. I’m on a lunch break at work typing this from my phone.

 

[levimax]

There is a large difference between these files apparently due to the distortion adding a bunch of HF content. No need to worry about "polarity inversion". My HF hearing is not that great and I could still hear a difference with speakers. With headphone I think I could get close to 100% on ABX but it is unenjoyable enough with music, with tones it is more like punishment.

 

[Thomas_A]

I have been very busy working today, so no more from me. These signals also seems to make my ears ring a bit...I do not want that...

If someone wants to hear an easy one it is this short part of a filtered sawtooth wave, first with correct polarity and then shifted to inverted polarity. Just say if you can or cannot hear the shift in timbre.

Dropbox

www.dropbox.com

 

[pma]

There is a large difference between these files apparently due to the distortion adding a bunch of HF content. No need to worry about "polarity inversion". My HF hearing is not that great and I could still hear a difference with speakers. With headphone I think I could get close to 100% on ABX but it is unenjoyable enough with music, with tones it is more like punishment.

Audacity shows the spectra wrong. I have it as well and have already checked it. Spectrum analysis in Audacity shows wrong level of individual tones in the multitone and also of the noise floor. It is not a professional tool for spectrum analysis. If you would like to see the spectra correct, please look here.
If you have REW or Adobe Audition, they will show you spectra properly. You might also use Deltawave or Multitone software by @pkane .

Thank you for listening to the files, but I need an abx report, otherwise the sighted impression does not help.

 

[levimax]

Audacity shows the spectra wrong. I have it as well and have already checked it. Spectrum analysis in Audacity shows wrong level of individual tones in the multitone and also of the noise floor. It is not a professional tool for spectrum analysis. If you would like to see the spectra correct, please look here.
If you have REW or Adobe Audition, they will show you spectra properly. You might also use Deltawave or Multitone software by @pkane .

Thank you for listening to the files, but I need an abx report, otherwise the sighted impression does not help.

I did a frequency analysis in Audition, green is original, red is BT and they look a lot different to me. I don't have experience looking at graphs of tones and then listening, I usually do it with different masterings, so I have no idea if or how audible these large visible difference are. Since the difference are down 40 dB to 50 dB I don't think just one difference would be very audible, I just don't know how "multiple differences" add up.

 

[amirm]

I offered up to 3k before. that's not the issue.

I missed this. If you are serious, please start a conversation and we will take it from there.

 

[pma]

I did a frequency analysis in Audition, green is original, red is BT and they look a lot different to me.

This is almost correct, regarding the amplitude drop of 15.9kHz and 20kHz spectral lines of the BT spectrum, except for the window and resolution used. But this below is what you should see if you set the analysis parameters properly:

And remeber that we listen by ears and not by eyes. So, regardless measurements, it is the listening test result that counts.

 

[krabapple]

Audacity shows the spectra wrong. I have it as well and have already checked it. Spectrum analysis in Audacity shows wrong level of individual tones in the multitone and also of the noise floor. It is not a professional tool for spectrum analysis. If you would like to see the spectra correct, please look here.

I never use a FFT 'size' that small (1024, the default) in Audacity. IIRC I use 8192 for Redbook audio (=4092 bins, time resolution 186 ms, frequency resolution 5 Hz)