|
Audio Asylum Thread Printer Get a view of an entire thread on one page |
For Sale Ads |
209.117.47.251
Hello everyone, it's up!
I know some people have strong feelings about digital upsampling filters and there's especially the notion of minimum phase filters being superior or more natural sounding than the typical linear filters used in most DACs.
Here's an opportunity to try it out for yourself and let me know if you hear a difference!
Note that the files are high-resolution 24/176 so please make sure your DAC is capable of this bit- and sample-rate.
Test will conclude on June 25th so please submit your results by that time.
Instructions and test download found in link below.
Thanks,
Archimago
-------
Archimago's Musings: A 'more objective' audiophile blog.
Follow Ups:
Thank you to all who participated!
Results out now in link below...
-------
Archimago's Musings : A 'more objective' audiophile blog.
Thanks guys for the submissions so far!
I see the responses and for the time being I think it's prudent to collect data rather than argue merits and details...
Remember though folks, digital filters represent one of the high-end "battlegrounds". A place where manufacturers try to differentiate from each other. And the pre-echo impulse response graph is one of those tools commonly brought out as a "concrete" example of the ills of digital audio.
This test I hope can at least look at one variable and provide some data on audible significance.
-------
Archimago's Musings : A 'more objective' audiophile blog.
Hi,
Your test makes no sense as what you are testing is not the upsampling filter, but it's interaction with the build in filters of the DAC the participant uses.
In other words, your test setup is absolutely incapable by design to make any determination on the subject you have chosen to examine.
I could possibly think of even more pointless endeavors, but working that hard gives me headaches.
Ciao T
At 20 bits, you are on the verge of dynamic range covering fly-farts-at-20-feet to untolerable pain. Really, what more could we need?
with facts?
Clearly, his objective is to produce his perception of a "scientific" test (never mind the details) and post the results as more *conclusive* evidence that high resolution audio is a waste of time.
Don't you get it by now about the Fairie Sorcerer? :)
It's even more complicated. There are at least three filters involved:
1. The filter(s) used to produce the 44.1 kHz input recording
2. The filter used to upsample 44.1 to 176.4 (as used to create the A file or the B file)
3. The filter(s) used in the DAC (e.g. digital and/or analog)
Your complaint is that the filters in (2) and (3) interact. This is correct, however this interaction is at the 176.4 sampling rate and above. As such, it is relatively less important than the interaction between the filters in (1) and filters in (2), which interaction is at 44.1 kHz. In any event, the test remains interesting if one's goal is to examine the difference between various upsampling filters one might use in a computer to upsample 4X before going to one's DAC. With some DAC chip architectures, entering the DAC at 176.4 bypasses two stages of sample rate doubling by the DAC chip, in effect replacing them with the filtering in the computer.
Doing critical listening is not a waste of time. One benefit is to learn how to hear certain subtle types of distortion. Being able to hear distortion is an important skill for a manufacturer, recording engineer, or audiophile. (Alas, the ability to hear subtle distortion may be detrimental to a music lover.)
Tony Lauck
"Diversity is the law of nature; no two entities in this universe are uniform." - P.R. Sarkar
Tony,
> It's even more complicated. There are at least three filters
> involved:
While that is true, two of them are fixed. The third may be anything though, from Non-OS via Craven Apodising minimum phase all the way to a standard "not quite brickwall" type.
> This is correct, however this interaction is at the 176.4 sampling
> rate and above.
Not quite. Look at the time domain and you find components much lower.
> Doing critical listening is not a waste of time.
I never said it was.
But if you desire to answer a given question with such test, you must construct the test in such a fashion that it answers this question precisely and you must exclude the possibilities of other interactions giving false positives/negatives.
The test here does not meet these criteria. As such it is a waste of time, just like 95% of those the ABX crowd usually inflicts upon us (I actually did a meta analysis of all published ABX Null results I could find - this suggested with substantial statistical significance that in the whole "meta-analysed" dataset of "null results" reliably detectable differences were present, just the dataset used and the conditions set were adverse to detecting them).
Ciao T
At 20 bits, you are on the verge of dynamic range covering fly-farts-at-20-feet to untolerable pain. Really, what more could we need?
As to those DAC filters. The A and B recordings are heavily apodized, since they were the result of upsampling four times. As a result, all good linear phase DAC filters are likely to produce similar results on any given file, i.e. the choice of these filters won't matter.
As a demonstration, I took the Piano A file and upsampled two different ways to 352/32fp using iZotope RX4. Both filters used pre-ringing 1.0 (linear phase) and filter-offset 1.0 (half band). The difference was in the filter steepness. One filter had steepness 32 and the other steepness 1020. (The later filter has a much longer impulse response.)
After generating the two upsampled files I created a difference file by mixing the two out of phase. The largest sample in this difference file was at -144 dBfs, which happens to be the level of roundoff error with 32 bit floating point files. Not much happening in the time domain.
On casual listening all three music files sounded the same to me and the difference file sounded completely silent. I tried boosting the difference file by 120dB (!). It was still inaudible and this surprised me. A spectrum plot of the boosted difference file showed that there was a peak of -65 dBfs at 88.4 kHz, which corresponds to a difference at -185dBfs in the unboosted difference file. There were also time variance in the background level that appeared to be correlated to the loudness of the original piano playing, but the boosted difference file had this time varying noise at even lower levels. (I would guess this has to do with the use of 32 bit floating point to represent the resampled files.)
Tony Lauck
"Diversity is the law of nature; no two entities in this universe are uniform." - P.R. Sarkar
As such it is a waste of time, just like 95% of those the ABX crowd usually inflicts upon us
Quite - but what tempts you to exempt the other five per cent? I've yet to see an audio-related "ABX" test that would pass muster in a school-level psychology course (after 50-odd years of observing the audio scene).
I actually did a meta analysis of all published ABX Null results I could find
My advice? Try to get out more.
D
Hi,
> Quite - but what tempts you to exempt the other five per cent?
There are rare cases where an ABX or ABX equivalent protocol has been used in blind testing where people did it right. Those tests rarely if ever see wide publication.
> I've yet to see an audio-related "ABX" test that would pass muster
> in a school-level psychology course (after 50-odd years of observing
> the audio scene).
I have seen very few.
> My advice? Try to get out more.
We have many rainy days in England... ;-)
Ciao T
At 20 bits, you are on the verge of dynamic range covering fly-farts-at-20-feet to untolerable pain. Really, what more could we need?
There are rare cases where an ABX or ABX equivalent protocol has been used in blind testing where people did it right. Those tests rarely if ever see wide publication.
Ah. The secret ABX-like test. Hadn't thought of that one . . .
We have many rainy days in England... ;-)
Hmmmm. Ever been to the west of Scotland?
D
Hi,
> Ah. The secret ABX-like test. Hadn't thought of that one . . .
Not so much secret as not particularly relevant to tales of cables etc...
> Hmmmm. Ever been to the west of Scotland?
Yes. Lovely place,except for the weather.
Ciao T
At 20 bits, you are on the verge of dynamic range covering fly-farts-at-20-feet to untolerable pain. Really, what more could we need?
You guys have no idea what a rainy climate is like until you live in Southern California. It just never stops raining here.
If you sell your Wilsons you can use the money to buy a lot of Nestle bottled water and use it to to keep your personal desert green. :-)
Tony Lauck
"Diversity is the law of nature; no two entities in this universe are uniform." - P.R. Sarkar
.
"Yes. Lovely place,except for the weather."
:)
Hi and thanks and just a question
Is it possible to replicate the test also with 16/44.1 files ?
Thanks a lot
Kind regards,
bg
Edits: 04/26/15
I would like the opportunity to compare your upsampling to my DAC's own native filtering.
Also, I noticed you downsampled two of the files from 88.2k. What kind of filter was used for downsampling? Assuming a linear phase brick wall type, do you think that would have any effect on the upsampling filter preference?
I am sure that I am about to say something stupid but how can I listen to these samples without hearing them via the filter in my DAC? That would seem to negate the exercise.
Edits: 04/26/15
He misses lots of things. :)
Your DAC sees 176.4k, so it's reconstruction filter is operating at 4x the frequency of Archimago's oversampling filters. That should be enough frequency separation so that your DAC's filter doesn't mask the differences between the files.
Welcome to Archimago's world.
Listening to DACs with multiple filters can still result in different preferences for a Minimum phase or Linear phase filter. It will depend on the sound of the DAC, your system, and ultimately your sonic preferences.
The Bricasti M-1 and the Resonessence Labs Invicta Mirus offered multiple choices when I reviewed these DACs.
I tried comparing the files as suggested (176.4 to Mytek) and with upsampling and conversion to DSD128 by HQPlayer with various settings: poly-sinc-short vs. poly-sinc-short-mp and DSD7 vs. DSD5.1. The various software settings each sounded slightly different, but my preliminary take is that the essential differences between the A's and the B's remain. This is consistent with the fact that filter settings at 44.1 are much more critical than filter settings at 176.4.
These are nice sounding clips, but perhaps not ones to place maximum stress on the filters as they don't have much high frequency content above 18 kHz. I suspect that brass or cymbals might make the differences less subtle.
Tony Lauck
"Diversity is the law of nature; no two entities in this universe are uniform." - P.R. Sarkar
Yes Tony, I agree high end is key here.
Yes, I understand that different filters will produce different sounds and therefore preferences. My own DAC offers a choice of 6 filters at the sample rate of the test files.
However for me to hear these files, whatever the filter chosen for their origination, I will have to listen to them via one of the 6 filters that I choose in my DAC for their replay. Others will either have to make do with the fixed filter that their DAC offers or, if provided with a choice, select their own combimation of file and replay filter. In all cases the filter chosen for the sample file can only be heard via another filter which may or may not be subjectivley compatible with the test filter. As one can only submit a prefernce for a sample without providing information as to the replay filter used in combination I cannot understand what useful information this test will reveal.
I hope that this clarifies my question.
Post a Followup:
FAQ |
Post a Message! |
Forgot Password? |
|
||||||||||||||
|
This post is made possible by the generous support of people like you and our sponsors: