Which is interesting; am i supposed to know if the question (and conclusion) is directed to me and even more disturbing do we really know what "that" is?
You made my point for me, precise stating of all conditions is important. Different hardware can mean anything, the claim stated (in fact here just this week) is that the same bits played through exactly the same hardware chain have an audible difference. If you play the files through different DAC's or clock recovery process you are no longer testing a single hypothesis. You should understand by now the claims are made in many cases deliberately to say our basic knowledge of the physical world at the most fundamental level is wrong (at least for hi end audio).
Last edited:
That´s truly circular logic.
Please explain how two bitwise identical files, played through exactly the same hardware, can sound different.
You are either trolling, or a Nobel prize is around the corner for you.
This is all off the track, we first need a mutually agreed upon method to determine actual audibility and then we can test any claims. I mention the extraordinary claims only because using them IMO is a distraction and waste of time. Let's stick with various DAC implementations or something similar.
You know that such thing will never happen because of those who are financially vested in boutique DACs, preamps, amps, cables & ...etc making audible difference.
What kind of evidence do you consider as sufficient?
That´s the link to my post, for simplification i copy your assertion and my response:
Waly wrote:
my response:
Obviously neither in your nor in my sentences there is "exactly the same hardware" mentioned, to the contrary we both wrote explicitely about _different_ hardware.
Or there is a third option that you´ve missed......
So we do agree at this point.
Seems that i´ve missed that statement so can´t comment.....
Even the same DAC but data delivery by different processes might be sufficient to evoke differences, that´s the reason why i/we often point to the distinction between claims of audibility and offered explanation for this claimed audibility.
I understand that, but to use always/mainly the most outrageous examples is a familiar usage of the eristics playbook (consciously or nonconsciously) and we have seen in this thread and the other recent threads about controlled listening tests, that even providing scientific evidence for some pitfalls of the commonly used tests (specific and in general) is blamed as "discrediting blind tests" .
So, as you´ve already suggested, it is much better to avoid these (often mainly bias driven) characterizations and to work out acceptable conditions for tests.
Your BS filter most likely automatically kicked it. It's one of Dan's (Max Headroom's) favourite "hypotheses"
Ah, no "that" is not all i´ve learned about....and yes, you could have saved me a lot of time if you had told me that 30 years ago (when we engineering students all developed a deep understanding of statistics without realizing it
) and even today it would help. (Jakob´s rule, a fact has to be repeated in audio forums at least 148 times before it slowly begins to sink in)Further, never mind, there are a lot more mysterious "thats" disturbing me.
Different packeting of USB data, for example. Of course the hardware must be "bad enough" for easy to measure effects of modulation of the clocks, supplies/references, etc, aka Logic Induced Distortion.
With synchronuous averaging and differential testing a lot of things can be isolated. Attached is a screenshot of an extreme example of this distortion. DUT was a lowly TASCAM US122-Mk2 bus-powered DAC/ADC which is low-fi enough to produce lots of this distortion. The setup was analog loopback by necessity (sample sync'd record while playback) so it's not clear if the DAC section is mainly responsible but actually it doesn't matter for the topic where at the DAC-->ADC conversion the disturbance occurs.
Stimulus was a LogSweep with a short pause with a 2^N length, repeated forever. Two recordings with 10000 averages were made which differed only in start offset of the sample buffers transmitted. Then condensed data blocks were subtracted after sample offset adjustment, with a level fine trim to get the lowest residual in the midrange frequencies (there is gain difference from slow reference drift as well less perfect nulling at the frequency extremes from slow clock drifts which cause small phase changes from all the analog lowpasses and high passes involved which spoils the nulling). Of course all calculation is done in 64bit-floating point, results were double checked and a lot of other experiments and null tests have been made to make sure the difference is real.
If there were no explicit distortion mechanism while the small gain and clock drifts would still be present the residual would not null perfectly as well, but the residual would be linear, only showing signs of random noise. As we can see, there is gross systematic distortion showing in the difference.
With the same technique I could isolate the microphonic/vibrational crosstalk into the device in a clarity that is breathtaking.
In the meantime I've refined the process so as to get rid of the influence of the gain and clock drift (which is always present even in state-of-the-art products) by interleaving the source data. Processing has also been optimized.
My best DAC/ADC, a RME ADI2-Pro FS, a much better device than the lowly TASCAM, now allows me to isolate effects to degree that keeps amazing me, both of the device itself as well as external DUT -- the loopback cable being the simplest DUT, isolating its microphony for example. Currently I'm in the process of designing and building an intelligent µC-controlled synchronous "switching box" so that I can actually switch something in hardware.
Attachments
Although the abstraction level seems appropriate to address the question, it was one aim to learn if beside the scientific requirements there are additional demands like "no financial interests involved" or "supervising" play a role.
Additionally the (still quite new especially in the audio field) demand for a preregistration - to avoid testing til finally some positive result appears which is then reported without mentioning the other attempts - or the qualification as "extraordinary" claim.
Usually we can be sure that it is impossible to control all variables, some can be controlled, the impact of others can often only be randomized and it depends on the objective if total control would be necessary or even desirable.
So on this level of abstraction i´d say the methodology has to follow best practice considering the objective of the experiment.
The question of statistical significance is always debatable as highly subjective, the usual 0.05 criterion should be sufficient as we would like to see replications anyway.
Yeah, like I could not imagine that playing the files through different hardware could produce different results.
This gets boring, it’s like fighting jello.
Can’t follow, but USB, even in different versions, won’t packetize data different, leading to different outputs. Let me know next time you copy a piece of software through USB of whatever version and the file(s) gets corrupted in the process.
As the person who made the pre-registration suggestion, I have no problems with a corporate sponsor although I'd be surprised to see any forthcoming (just to be clear, not saying you suggested I did). The p <0.05 wouldn't be my first suggestion as many fields are moving away from specific endpoints as "valid" since it creates an effect of chasing validity rather than pursuing knowledge. AIP holds onto its 6 sigma for discoveries but the efforts involved to get that are, well, CERN like in nature.
We can argue significance until the cows come home after the data is out
- Status
- Not open for further replies.