DAC audible differences are finally proven with ABX?

[KSTR]

2. Polarity reversal. I have never investigated this. I imagine people here who say it could be audible say so in the context of one DAC having reverse polarity vs another in ABX, not by itself. Correct me if you are saying polarity reversal by itself is an audible impairment in which case, I like to see some evidence.

For most recordings it is impossible to define a correct polarity but for some it is, those are made with a single, or single stereo set, of microphones. Some of those recordings are released with actually correct polarity, other are inverted (there are list online where one can check which label / which recording inverts). Instruments with highly asymmetric waveforms like trombones and trumpets tend to have slightly different timbre when inverted. It will need a very experienced listener/musician with decades of exposure and training who can tell if a trombone is inverted right away. Same goes for the sound of drums etc.

Regardless of the above, IME and IMHO for any given recording there are typically two aspects of a polarity switch that people can perceive (with linear phase loudspeakers, much preferred over headphones), one is about "bass/midrange authority" and the other is "soundstage precision". If you are lucky, one polarity setting has more authority of the low fundamentals and a more pin-point soundstage at the same time, more often you get one but not the other. On top of that, some people may prefer a more diffuse soundstage which may sound "more 3D".

Bottom line: except for rare corner cases a wrong acoustic polarity will result in relevant impairment of the reproduction.

 

[pma]

Some of those recordings are released with actually correct polarity, other are inverted (there are list online where one can check which label / which recording inverts)

The famous infamous Telarc 1812 (cannon shots). Disc data and the recorded data (from power amp output):

 

[firedog]

And there isn't on AS ?

Sure, but they don't proclaim they're a science site.

 

[amirm]

That's a nice "unscientific" approach. You know very well that those numbers have nothing to do with actual visits - they include bots. And how many visitors at each site have tracking blocking software installed? Not reflected here.

Similarweb uses an array of tools to sample website traffic including actual browser plug-ins of actual users. Same metric is applied to us that is applied to AS. So while the numbers are estimated, they are estimated similarly for both of our sites, showing us to be nearly 9 times more popular. In other words, AS is round off error to our level of popularity.

For grins, this is what ChatGPT Pro thinks of the value of our sites:

Factor	Audio Science Review	Audiophile Style
Traffic/reach	Much higher	Lower
Forum/community asset	Very strong	Strong
Search authority	Very strong	Good
Proprietary reference content	Exceptional measurements/reviews	Good editorial archive
Brand differentiation	Very high	Moderate-high
Owner dependence	High	High
Estimated public-data value	$1.5–3M	$300K–800K
Likely ratio	~3–5× AS value	1×

As a general comment on the thread, it's certainly possible there were "tells" in the test that Josh unknowingly heard. But I find it interesting that some of the "tells" being discussed here are things that in other threads have been dismissed as inaudible. Well, are they possible tells or are they inaudible? Are they inaudible when you agree with the results and audible when you don't?

It is all speculation because Josh has left of us starving for facts. Nothing is assumed to be the actual reason for audible differences; only potentials to investigate assuming no tells. I have not talked about any of that because the most fundamental issue is still the comparator. It could have imparted differences that are far, far more substantial than any other ideas.

 

[firedog]

For grins, this is what ChatGPT Pro thinks of the value of our sites:

Good you said "for grins", because the Chat is nortouriously unreliable when asked for those kinds of evaluations.
It regularly gives false invented results.

 

[amirm]

There's some entertainment "science" going on here. But hey, that gets those clicks, doesn't it.

With a site as large us, and so many substantive topics in parallel, this thread is nothing. For AS though, it is probably a big deal especially since we are sending them traffic/clicks. So hopefully I get a birthday card from them when that day comes. And remember, we don't monetize clicks or any kind of traffic here.

 

[Deleted member 93733]

Sure, but they don't proclaim they're a science site.

So are you saying that as Josh’s work was posted on that site, it is also non-science?

 

[amirm]

Good you said "for grins", because the Chat is nortouriously unreliable when asked for those kinds of evaluations.
It regularly gives false invented results.

Not if you read and understand what it presented (a requirement for use of AI).

Here is Gemini's take on it:

AudioScienceReview.com (ASR) commands a significantly higher valuation than AudioPhileStyle.com (APS), primarily driven by an exponential gap in web traffic, global reach, and community engagement.

Because both entities operate as closely held, private digital communities rather than public companies, their precise enterprise valuations are not public. However, by treating them as digital media properties and applying standard industry multiples (typically 2.0x to 4.5x annual net revenue based on traffic, programmatic ad yield, and sponsorships), we can establish a reliable baseline comparison.

Executive Comparison

Valuation Driver	AudioScienceReview.com (ASR)	AudioPhileStyle.com (APS)
Estimated Value Range	$1,500,000 – $3,500,000+	$150,000 – $350,000
Traffic Tier	High (Multi-million monthly pageviews)	Niche / Boutique
Primary Philosophy	Strictly Objective (Data / Measurements)	Subjective & Digital Audio Focus
Core Monetization	Member donations, Tiered Sponsorships	Display ads, Sponsored Content, Affiliate
Community Health	Hyper-active, fast indexing thread growth	Stable, slow-growth legacy forum

1. Analyze Traffic and Audience Metrics
Traffic is the primary currency for website valuation. ASR dwarfs APS across every measurable user metric, positioning it to command a 10x premium in baseline traffic asset value.

• • Global Visibility: ASR ranks among the top audio authority platforms globally. It serves as a mainstream hub for consumers looking to validate hardware purchases using Audio Precision analyzer charts.
  • Long-Tail Search SEO: Because ASR publishes objective, standardized hardware measurements (SINAD, frequency response, linearity charts), its review threads index aggressively on Google. This creates an evergreen flow of high-value search intent traffic.
  • Niche Contraction: APS (formerly Computer Audiophile) holds historical value as a pioneer in digital audio streaming and server optimization. However, its traffic footprint has consolidated into a much tighter, boutique demographic of legacy audio hobbyists.

2. Evaluate Monetization Strategies
The structural valuation of both sites is dictated by how they choose to monetize their respective audiences.

• • AudioScienceReview's "Pure-Play" Donation Model: ASR consciously limits aggressive display advertising to preserve the perceived integrity of its scientific testing. It relies heavily on community donations, patron tiers, and manufacturer equipment loan programs. While this reduces programmatic ad yield, it builds immense brand equity and user loyalty, which mitigates churn and stabilizes long-term asset value.
  • AudioPhileStyle's Commercial Model: APS relies heavily on traditional ad banners, industry sponsorships, and affiliate links. Because its audience skews toward high-net-worth individuals willing to spend five figures on bespoke servers and cables, its Average Revenue Per User (ARPU) from direct industry advertisers is exceptionally high. This keeps the site profitable despite lower raw traffic counts.

3. Assess Intangible Brand Value & Moats
A digital media acquisition relies heavily on defensive moats—how easy is the community to displace?

• • The ASR Data Moat: Founded by Amir Majidimehr, ASR has cataloged thousands of proprietary hardware benchmarks. This database cannot be easily replicated by a competitor without millions of dollars in test gear (such as the Klippel Near-Field Scanner) and years of labor. This data archive acts as a massive intellectual property asset.
  • The APS Community Legacy: APS's value lies in its forum archive, which features deep historical knowledge on digital-to-analog conversion, network filtering, and specialized streaming protocols. While highly valuable to an audiophile purist, it lacks the broader consumer appliance appeal that drives mainstream valuation multipliers.

 

[antcollinet]

Bottom line: except for rare corner cases a wrong acoustic polarity will result in relevant impairment of the reproduction.

Should there be a "not" after the "will" in that line? As written, it seems to contradict the non bottom lines

 

[RandomEar]

Who says there is an active preamp? Increasingly there aren’t. The ABX comparator is exactly the sort of device that would hi light this issue. There are real world amps with lower than ideal input impedance. Does a defined transparent DAC have to deal with eg tube amps etc or capacitative outputs etc?

I’m not saying that these effects aren’t measurable, of course they are, but do you have?


Are these detailed measurements done for DACs which have been deemed “transparent”?

Again I’m not saying that these measurements can’t be done but I didn’t see that as part of the definition of “transparent” … I’m looking for edge cases. I don’t want to miss simple explanations

But how his this relevant to the case discussed here? The HPA4 used by Josh has an input impedance of >>50 kΩ on both, unbalanced and balanced inputs.

 

[RandomEar]

I haven't seen the source code, but my guess is that it can estimate a clock drift once it has aligned the start of both samples. To compensate, it needs to do a Dynamic Time Warping, which in turn means aligning any data to fit the source over the given a given window (say 4ppm). This means that I can move freely in that window, but unfortunately would eliminate actual differences in the time domain.

That together with gain alignment, one could ask if we are actually comparing fairly with DeltaWave, since we adjust both axes at the same time to fit the source.

"Dynamic time warping" sounds like Star Trek. It does a simple linear drift correction if you let it. It's active by default. Of course this has an effect on the much toted time domain, because it needs to correct increasing time differences in both recordings. They are not audible when you listen to each DAC as is, but they would destroy any hope of a fair comparison in an ABX test if you dont just listen to the first seconds of a track.

There is also an option for a non-linear correction in DeltaWave, which is off by default. I have used it and the results look interesting, but for the cases I have seen it did not lead to relevant improvements in the measured numbers. Probably because the samples were short-ish (<1 min). It's also not a very fine grained correction, which would result in weird "warbling" changes on the millisecond scale: It looks like some form of smoothed piece-wise polynomial or spline.

In any case, none of these settings work on a per-sample basis in the sense that each one is aligned independently. It does not invalidate the comparison.

 

[pma]

Careful there. Lower input impedance in the pre-amp brings down the levels but in a uniform, frequency independent way.

Amir, would you mind making a comparison of 100k/600ohm load impedance at 1kHz, not at 50Hz? I am usually getting rise of high order distortion components in such test. But not at 50Hz, as the feedback factor is usually higher at 50Hz than at 1kHz

I have to correct my point above, with now available good DACs the load impedance does not seem to be an issue. I have compared a record with Cosmos (2.7V range) to the record when my input buffer with 100k input impedance is before the Cosmos. There is almost no difference.

But the problem with LF drop and rising LF distortion of the Cosmos remains the same, when we try to compare recorded vs. original file.

 

[pma]

"Dynamic time warping" sounds like Star Trek

https://www.researchgate.net/publication/285279006_Dynamic_time_warping

 

[fivepast8]

Hi Josh,

I am ok with the offer $3k: $1.5k to you, 1.5k$ to charity of your choice, plus shipping costs of the unit to @amirm, provided Amir is ok to receive and measure it.

But since I am in Switzerland, it is less easy. I won’t be able to deduct the donation, which would allow me to reduce my taxes by roughly 30-40% of the donation, ie $500 minimum. I’m not sure this is the same range in the US.

So, to make things simpler, can we agree that I send $2.5k+shipping to you and you donate as you want?

Else, I can donate $1.5k to the equivalent organization in Switzerland that I’m member of:
WSCHWEIZER TIERSCHUTZ STS. But they will take up to 2 months to confirm receipt, so that’s cumbersome, and even if I could produce proof of payment, I also could retract and it is not what you want.

I’ll DM Amir too, to check if he’s ok to support the receipt of the device and measurements.

Regards

Flo

Hello Flo, I am happy to contribute $400.- towards the purchase / measurement / clarification and advancement of science. I will send the money to you by TWINT and in CHF and hopefully some others will follow with some donations as well. Should the box be sold for a profit later... I want my share as well !

 

[NTTY]

I have to correct my point above, with now available good DACs the load impedance does not seem to be an issue. I have compared a record with Cosmos (2.7V range) to the record when my input buffer with 100k input impedance is before the Cosmos. There is almost no difference.

View attachment 550350

But the problem with LF drop and rising LF distortion of the Cosmos remains the same, when we try to compare recorded vs. original file.

View attachment 550351

I was about to propose running the test, but you’re way faster than me!
All the above being said, if you use the Scaler in front of the Cosmos, as the high impedance input buffer, that solves this issue, right?

Also, when it comes to recording and analog signal, we need to think about what headroom we want to have, to prevent unwanted overs. And any headroom we give to the ADC, means almost the equivalent loss on DR of the DAC under test. Say we leave 6dB headroom for the ADC, and it is no longer possible to claim we get the full SINAD of the high-precision DAC.

To achieve measurements Amir show on SINAD, it is necessary to use an analog notch. From memory, with the best DAC that I currently have, I think I measured 127dB SINAD with the notch and 123dB without. If I add 6dB headroom at the ADC input, I suspect the same measurement would show 118dB or so.

This alone would equalize the performances between DACs, just because the ADC is inferior to most of them under these technical requirements for recording.

So again, to me, it is everything but easy to record high precision DACs, as you said.

 

[pma]

All the above being said, if you use the Scaler in front of the Cosmos, as the high impedance input buffer, that solves this issue, right?

I am not sure. The high input impedance of the buffer or the Scaler avoids loading of DAC output, but does not help with Cosmos electrolytic coupling capacitor value and LF distortion. This remains unchanged. Some cure is if you use higher Cosmos input range or series resistor with the input, for obvious reasons (capacitor CR HPF corner and distortion). But, then you lose resolution (input range sensitivity).

I am in fact not much interested in getting the lowest 1kHz SINAD. This is just good for the chart rating at the single frequency. But as you, me et al can see, we are fighting different issues. Namely FR flatness, ringing, LF phase behaviour. They all may become audible. NOT the 1kHz SINAD. No difference if it is -110dB or -130dB. Just numbers game and top chart race. This is interesting for producers to make a techno oriented marketing.
-----------------------
To make it clear - I consider E1DA Cosmos ADC an excellent product, especially concerning its price/parameters value. But it has certain limitations and we are here not to create a cult, but to speak openly.

 

[RandomEar]

https://www.researchgate.net/publication/285279006_Dynamic_time_warping

I hate that it's actually a defined term Next you'll probably tell me transparent alumium is now a thing... hey, wait a second!

(Thanks for linking it)

 

[NTTY]

Hello Flo, I am happy to contribute $400.- towards the purchase / measurement / clarification and advancement of science. I will send the money to you by TWINT and in CHF and hopefully some others will follow with some donations as well. Should the box be sold for a profit later... I want my share as well !

Thank you! This is really kind of you.
Josh promised me he will donate to his local SPCA after refunding himself on the initial total cost of the box. So you’re indirectly contributing to better animal protection in USA

 

[NTTY]

I am not sure. The high input impedance of the buffer or the Scaler avoids loading of DAC output, but does not help with Cosmos electrolytic coupling capacitor value and LF distortion. This remains unchanged. Some cure is if you use higher Cosmos input range or series resistor with the input, for obvious reasons (capacitor CR HPF corner and distortion). But, then you lose resolution (input range sensitivity).

I use mine at 10Vrms always feeding it from the scaler. With a 4Vrms output from the DAC (as Amir does) and the scaler set at +7db (first step), I’m close enough to the max of the ADC.

I indeed reported some low level added distortion from the ADC at low and high frequencies, when the signal is close to full scale (on the review of the SMSL DO400). But we are talking -110dBr in areas where our ears are not very sensitive to such.

Ok, enough of deviation, I promise. All that was to say it’s not straight forward to record from the ADC, but it can be done with care.

 

[pma]

Yes at 10V Cosmos makes the best job, as the series resistor in the input is highest. It is good for both LF response and distortion.