Hi-Res - Does it matter? Blind Test by Mark Waldrep

[Miska]

6 hours ago, Ajax said:

I will be doing a thorough analysis of each file and providing the spectra and dynamic analysis to participants. I've already done that for a test file by The Latin Jazz Trio. Here's the spectra of all of the formats:

The Spectra of "Memories of Rio" in all six formats

 

One of the spectrum plots where there is clearly something wrong. Spectrum goes straight to the Nyquist of 96k sample rate. I guess yet another ADC that has got digital decimation filters wrong.

 

Have to download and see how many of these contain some actual content in the higher frequencies.

 

 

P.S. Never mind, doesn't seem to be actually available for download.

 

[Miska]

4 minutes ago, kirkmc said:

Also, the fact that the 256 kbps MP3 has higher frequencies than the Redbook is a bit odd. Something wrong with the measurements? 

 

I think it is 48 kHz sample rate MP3.

 

Rest of course depends on the particular MP3 encoder. Not all the encoders sound or work the same.

 

5 minutes ago, kirkmc said:

This test should include AAC at 256 kbps as well.

 

Yes, that is interesting too. To me personally, AAC sounds better than MP3, and I think I know the technical reason for that too. Of course in that case too, a bit depending on the particular encoder implementation.

 

[Miska]

1 hour ago, Ajax said:

That is why I am encouraging everyone to participate in Mark's study. The bigger the sample base the more reliable the results.

 

I guess it is quite hard to get enough diversity in participants that it would be statistically valid for entire earth's population.

 

I rather take safe approach that recording should capture everything that can be captured of the performance, in terms of dynamic range and frequency spectrum. This also includes microphones, thus using mics like the Sanken's 100 kHz bandwidth one. And then ADC should record it all without removing anything. Regarding audio, this is not a problem at all with current network bandwidths and storage capacities. My 60 fps 4K video recordings are much more resource hogs in all shape and form, and soon things will move over to 8K and probably 120 fps.

 

[Miska]

One thing good to note about these type of comparisons is that listening to the file that has been converted to 44.1/16 and then back to 96/24 in software is not the same as sending the 44.1/16 data to the DAC...

 

[Miska]

4 hours ago, Confused said:

Yes, a good point.  However, let's say you have a set up where you typically upsample everything (including 16/44) to 24/192, would this eliminate the difference, either in part, or maybe 100%?

 

There is still difference between upsampling filters. The one you use for upsampling, and the one used to convert files. Another factor is of course conversion algorithm used to convert the original to 44.1/16 first.

 

So what you hear in this case for "44.1/16" source, is not representative of what you would hear in the case of true 44.1/16 version of the same track.

 

You could also have some fun and do those same conversions with various different pieces of software and compare. Results are not the same.

 

I personally would prefer to buy the original recording format. That of course allows one to convert it to 44.1/16 for listening if wanted (I don't know why), but would also future proof the content and allow re-conversion with another algorithm at later time. Instead of being stuck one second generation version. This is also one reason Apple has been asking labels / music producers to deliver their content in 96/24 format for a veery long time, because they can then re-convert it as necessary if they decide to change the delivery format. If they want to switch to HiRes, they can pretty much just flip a switch.

 

 

[Miska]

1 hour ago, Confused said:

So thinking about this, if we are told what filters are used to convert (upsample and down-sample) the files used for Mark Waldrep's "blind" comparison test, would it be possible to come up with an optimum filter(s) in HQPlayer to perform the comparison?  I am wondering if there is a possibility to have a setting in HQPlayer that would make this test as valid as possible?  Or maybe this is not possible?

 

Since they are all back to same format and already upconverted, you can compare with the caveats I mentioned. You can use whatever HQPlayer settings you would normally use.

 

I just wonder why the 44.1k version is not left as 44.1k version, it would make the comparison more representative when listening is done with a system that would be anyway used for listening similar files.

 

[Miska]

24 minutes ago, audiobomber said:

Decreased potential for confirmation bias?

 

Why would the file format make any difference to that? It takes about two seconds to check which file is which...

 

[Miska]

1 minute ago, audiobomber said:

I have a DAC that indicates sample frequency and bit depth, and another that shows sample frequency with coloured LED's. I would not deliberately check bit depth, because that would defeat the point of a blind test, but I could spoil the test inadvertently.

 

Yeah, mine too. It says constantly "DSD256"...  Regardless of source content.

 

The results would be more representative of reality if 44.1k would stay 44.1k, without excuses. If you don't want to see the display, you can put couple of PostIt notes on it.

 

[Miska]

7 minutes ago, audiobomber said:

I will be able to tell whether his claim that filtering eliminates the need for hi-rez recordings.

 

Ehm, what claim is that and where? What I've read this is purely yet another RedBook vs HiRes test. I've lost count how many there have been before.

 

[Miska]

41 minutes ago, Ralf11 said:

 the Reiss meta-analysis is the the paper to read

 

Ehh?

 

[Miska]

2 minutes ago, Jud said:

http://www.aes.org/e-lib/browse.cfm?elib=18296

 

I'm still missing the context how that is related to Mark claiming something about filtering? Apart from pondering about rate conversions and filtering inherent to those processes (not explicitly mentioned).

 

I cannot see anything like that here: http://www.realhd-audio.com/?p=6197

 

[Miska]

7 hours ago, Ajax said:

This tone can reach 3.01 dB over 0 dBFS before the maximum digital codes are reached.

 

With some tracks, intersample peaks can exceed that...

 

 

[Miska]

18 minutes ago, sandyk said:

 I have the original Vertigo 800 088-2 version made in Hanover W.Germany by Polygram and it sounds way better than any remaster that I have heard. It was manufactured in 1982.

 

Not surprising. Many new pressings are just created blindly from RedBook masters. While the old ones were mastered specifically for vinyl...

 

[Miska]

1 hour ago, gmgraves said:

I can compare both to the master DSD file. I know what the differences are, and through my Yiggy and the Chord HUGO and the Chord Quetest as well as the AudioQuest Cobalt.

 

Too bad none of those play DSD natively...

 

[Miska]

1 hour ago, Ralf11 said:

There are no experts in matters of taste.

 

I hope you are familiar with the very strong effects of confirmation bias on wine "experts" tasting perceptions...  Hint: the label is critical.

 

Not to say that there aren't real differences.  I really like that Romanee plonk.

 

I'm bad person, I don't value labels or brands. I value my taste. But certainly wines taste different. And so do whiskies and dark rums. Shiraz and Pinot Noir don't taste the same. And there's no way to objectively state that one wine is better than the other.

 

It is like one scientist not long ago stating in a magazine that "all salts taste the same". Meaning that all pure NaCl tastes the same. He continued adding that "impurities in some salts may make it taste different". He quite didn't get that those "impurities" were the essential difference between rock salt and sea salt. And that those were precisely the small but very important difference. And that the 99% of NaCl was the uninteresting boring and largely unimportant side detail.

 

[Miska]

4 hours ago, jabbr said:

The reason this is so important is that it provides a clear cut mechanism for the audibility of ultrasonics — not that you can hear, for example, a 30 kHz tone, rather that the full range sound of a cymbal might sound different than the 20 kHz stopband filtered recording of this cymbal. 

 

I'm personally along the lines of Oohashi research. If I take out just the ultrasonic part, I cannot hear anything. But I can hear difference between content that has ultrasonics removed vs content that has ultrasonics left intact. So the baseband part is fundamental, with it's extension.

 

My view is that hearing is not like FFT splitting frequencies apart and analyzing those independently of each other. Instead, it is more like wavelet analysis where things are matched against certain custom filters. For example things like transient leading edges. There frequency content defines shape of the initial attack wavelet. If you remove any part of it, it breaks down more or less; the shape doesn't match anymore. IOW, it is different to have separate uncorrelated discrete tones, vs having correlated combination of harmonics that define exact shape of a waveform.

 

There's also some research, that if you remove initial attack phase of instrument sound, and leave only the steady state part, it becomes much harder to detect the actual instruments. Thus the initial attack part of the signal is very fundamental.

[Miska]

18 hours ago, Ralf11 said:

it may not be the impurities but the grain size for salts

 

Just dissolve it in water or food and compare.

 

I rarely have non-dissolved grains of salt left in cooked food...

[Miska]

12 minutes ago, gmgraves said:

That ultrasonics modulate hearing is not grounded in certainty either. There are only two things about hi-res that are for certain: 1) that a very high sampling rate moves the Nyquist frequency further away from the audible pass band, and 2) 24 or 32 bit sampling allows for better low-volume resolution and/or more headroom.

 

Especially with DSD, you have Nyquist at minimum of 1.4 MHz...  And don't need to worry about number of bits, but instead get even better low level linearity.

 

13 minutes ago, gmgraves said:

This manifests it self in Hi-res downloads costing more than CD quality downloads. There is really no reason why this should be so except that audiophiles seem willing to pay $25 per hi-res album for the same music that is available at CD resolution for often less than half that! Never let it be said that retailers would deny folks the privilege of paying all that the market will bear.

 

Given bitrate factor of 3.3 between 44.1/16 and 96/24, I don't consider album price of 16,50€ bad for 96/24 midres.

 

[Miska]

Just now, Ralf11 said:

Miska - do you have a cite for the initial attack paper?

 

Sorry, I don't have that kind of database in my head. And I'm not trying to be librarian either. It goes more into music research side about instruments and less to about engineering sciences though.

 

[Miska]

3 minutes ago, crenca said:

Also, @esldudeposted some measurements on some thread (can't remember which one) where he looked for a "transient" or 'rapid rise'  of the usual suspects (a drum and a brass instrument if memory serves).  He could find none if I recall correctly.

 

I think this counts as fairly rapid rise:

http://www.cco.caltech.edu/~boyk/spectra/11.htm#b

 

I did bunch of my own recordings with various instruments and got pretty fast responses also for example from soprano glockenspiel.

 

To correctly reconstruct the exact spike waveform shape, you need a lot of high frequency harmonics.

 

For these reasons I've also started again to measure TIM distortion from playback gear using the standard method. Too bad it seems to have been forgotten for some decades.

[Miska]

48 minutes ago, esldude said:

Guys, there is some frequency beyond which it simply cannot matter to us humans.  Maybe it is a little more than 20 khz for some rare situations maybe it isn't.  Even people with good high frequency hearing have a steep, steep rise in the threshold once you pass 15 khz.  The idea 44 or 48 khz sampling drastically effects playback quality is quite ridiculous I think.  If you do everything else right in the chain, 44 khz recordings vs higher rates might be just barely audible as different.  It isn't like 44 khz will sound significantly degraded and 96 khz or higher will sound wonderful.  Of all the things involved in getting a good recording the sample rate ranks way down near the bottom of such a list, and may not matter at all.  

 

Just use so high sampling rate and bit depth that you have the analog noise floor dominating everywhere and that all harmonics captured by the microphones have long disappeared into the noise floor. And you are safe. And absolutely no reason not to do so.

 

Or use DSD and you don't need any brickwalls anywhere...

 

[Miska]

49 minutes ago, gmgraves said:

The Neumann, AKG, Telefunken, Sennheiser, and Sony condenser mikes as well as the dynamic varieties and contact mikes favored for pop and rock recording, and likely to be found in a professional recording studio’s complement, simply have little response above 20 KHz.

 

Like Sennheiser MKH-8020 for example that goes to 60 kHz? Or the DPA microphones like 4006 and especially 4007 that go to 40+ kHz? Or Sanken CO-100K that goes to 100 kHz, used for example by Five Four?

 

What I've looked, for example in recordings made by 2L, content goes up to about 60 kHz. So you need at least 120 kHz sampling rate for those.

 

[Miska]

1 hour ago, Rt66indierock said:

We've had high resolution for a couple of decades and nobody cares at the consumer level. And if you take out Japan does anyone care about DSD? 

 

Yes, quite a lot of people care about hires (PCM and DSD). All the time new material coming, to nativedsd.com, highresaudio.com and even HDtracks. Nowadays more are more at DSD256 rates.

 

I have two DSD256 capable AD/DA converters (RME ADI-2 Pro) and my own recording software. But of course many of the recordings are made using Merging, TASCAM, Korg or Mytek hardware and edited in Pyramix or Sonoma. Now Merging has three DSD256 capable recording devices (Horus, Hapi and Anubis). All those can do also playback, in addition to their NADAC.

 

[Miska]

2 hours ago, gmgraves said:

I have been capturing in DSD for over a decade, now, for that very reason. I rarely listen in DSD except on my computer. Ironically, except for my Oppo 205 (the analog output of which is not connected to my amplifier), my main system’s DACs do not support it. No biggie though; If I make copies for anyone (usually the group members, or the orchestra association for the use of the conductor as a “study CD”), I give them Redbook transfers from the master. I also make hi-res LPCM transfers for myself (or my clients, if they wish) which I copy to my NAS for listening on my main system.

 

I think I have only three DACs that don't support DSD. All the rest do support DSD, even up to DSD1024...

 

When I buy content I prefer to buy it in the original recording format, not some conversion.

 

I very rarely send LPCM to a DAC though.

 

[Miska]

6 minutes ago, Rt66indierock said:

They don't add up to anything but a rounding error in the market.

 

I only care that music important for me is available in hires. That ranges from Daft Punk to David Gilmour and Mark Knopfler, and to classical music and blues/jazz. I'm not even rounding error as customer for a record company, so I care as much about their market as they care about me.