Two unresolved issues

[Lee]

I find it interesting that we have two professional recording engineers with real studios and top notch recording gear that are in agreement with each other about what CD can't do compared to Hi-Rez and on the other side we have what I would call a home hobbyist with a very modest recording set up (and I'm being charitable here) who claims that redbook CD is perferct, it captures everything, and is always demanding "proof" from those who know better.

That is rather interesting. I cannot at all compare with Bruce if you are including me in the grouping. It's a nights/weekend but very serious second job for me. I'm sure Bruce has forgotten more than I will ever know. Maybe even forgotten more in one night than I will ever know.

I like Ethan's inquisitiveness but I don't think he will ever be satisfied with "proof" of what I hear. I can only hope that one day he can visit me in Atlanta and hear a session.

 

[Phelonious Ponk]

That is rather interesting. I cannot at all compare with Bruce if you are including me in the grouping. It's a nights/weekend but very serious second job for me. I'm sure Bruce has forgotten more than I will ever know. Maybe even forgotten more in one night than I will ever know.

I like Ethan's inquisitiveness but I don't think he will ever be satisfied with "proof" of what I hear. I can only hope that one day he can visit me in Atlanta and hear a session.

I know with absolute certainty that I've forgotten more than I'll ever know. I'm not nearly as sure about hi-res vs. redbook.

Tim

 

[Phelonious Ponk]

Tim,

I want to address these three points, since I think we are on the same page otherwise. First, I think SACD is a a far better medium than CD. It has a more pure midrange and the noise shaping that occurs in the HF region is so far out that I and others simply don't hear it. I've got a thousand SACDs and when they are good masterings. I also was at several of these sessions and know what the event sounded like. Even with reference level CD playback, the SACD on a good player (no PCM conversion, etc.) always wins versus CD. There is just something wonderful about DSD encoding. Also, if you record a string ensemble playing the DSD encoding will slightly edge out the 24/176 recording. Both will be really terrific but I prefer the DSD. Of course, the microphone placement and skill of the engineer is also critical. Some SACDs simply sound bad because of the poor original recording. You cannot polish the proverbial turd.

I'm not sure I follow the third point above on DBTs. My point is that having critical listening skills is paramount and many of these tests which are designed to test equipment differences really wind up revealing those that do and don't have critical listening skills. But it can be an important tool. Many high end manufacturers do both objective measurements and subjective listening sessions. I like that approach.

So if you ask me if I like an objective approach or a subjective one, I would answer "Yes". I think the truth lies with using both.

Let me clarify, Lee. The comparisons I've done were between hi-res downloads and the same files down-converted to redbook in Audio Midi Setup in OSX. It's not a particularly good test, because you can't switch back and forth between them quickly, so if the differences are subtle, they can pretty easily be missed. I missed them. YMMV.

I think the jury is still out on "critical listening skills" (which can mean a lot or very little) and DB/X testing. You can make the argument, as Sean Olive does, that trained listeners focus the testing and get you more valid results much faster. Or you can argue just as validly that trained listeners know what to listen for, therefore have expectations, therefore are much more subject to bias. The only thing I know with certainty is that better-designed tests, more samples, more trials = more valid results. It's fairly rare when the results of the first round of tests in a well-designed study changes with more trials, but statistically, you still need to exceed the margin for error or you really haven't satisfied anything but your own curiosity.

Tim

 

[Lee]

Let me clarify, Lee. The comparisons I've done were between hi-res downloads and the same files down-converted to redbook in Audio Midi Setup in OSX. It's not a particularly good test, because you can't switch back and forth between them quickly, so if the differences are subtle, they can pretty easily be missed. I missed them. YMMV.

Tim, what do you listen to in terms of playback?

 

[JackD201]

I pretty much agree with the second paragraph Tim. If I just might add one thing, is that for a conclusion to be accepted, the margin of error must be agreed upon. That margin may be reduced at some later time when needed. Migration from 16/44.1 to ever higher resolution is an example of this and while it is a continuing process, we won't get any finished products made if we don't peg things down at waypoints at least temporarily.

Philosophically I think this is where "What's good enough" and "Whats best - right now" really part ways hence the heat. Tolerances (margin for errors) are tighter for those pushing the boundaries. Skeptics will counter that the cost of achieving these tight tolerances are wasteful. They will also say (and there is a lot of truth to this) that there are individuals out there who attempt to cash in on the situation by passing off "bling" as superiorly engineered products. These are however two wholly separate issues. My personal observation is that what heats up those after better performance is when these two issues are lumped together because doing so is tantamount to accusations of fraud.

 

[Phelonious Ponk]

Tim, what do you listen to in terms of playback?

Mostly lossless files from a MacBook Pro.

Tim

 

[Lee]

Mostly lossless files from a MacBook Pro.

Tim

So you are comparing like FLAC 24/96 files versus redbook in lossless? I wonder if the Mac's soundcard would allow you the best DAC for the hirez.

The better the resolution of the playback system, the more noticeable the difference between 16/44 and 24/96.

 

[Stereoeditor]

Let me clarify, Lee. The comparisons I've done were between hi-res downloads and the same files down-converted to redbook in Audio Midi Setup in OSX. It's not a particularly good test, because you can't switch back and forth between them quickly, so if the differences are subtle, they can pretty easily be missed. I missed them. YMMV.

The Quicktime real-time sample-rate converter you are using is demonstrably not transparent - it's optimized for speed rather than quality - so I am puzzled why you did not hear any difference. A better test would be to prepare highest-quality downconversions using something like Bias Peak set to its best-quality or the inexpensive Wave Editor program, which includes the excellent iZotope SRC, and AB those Red Book versions against the originals.

John Atkinson
Editor, Stereophile

 

[Phelonious Ponk]

If I just might add one thing, is that for a conclusion to be accepted, the margin of error must be agreed upon.

Consensus is not how margin for error works, Lee, but that's a moot point. We would never get a margin for error agreed upon unless everyone agreed with the results.

I am puzzled why you did not hear any difference.

Don't be puzzled, John. The more I hang around here, the more I realize I'm much more of a "what works for me" than a "what's best" guy. It's not that I don't appreciate really good recordings and really good gear. I absolutely do. And I have pretty good ears for subtleties in recording technique, subtle differences between the sound signatures of instruments of the same category...I can definitely hear the difference between digital and analog recordings and even digital and analog media. But some of the stuff you guys talk about, I just don't have the patience to hear. I start out pretty good. I concentrate really hard and try to pick up on the length of the reverb trails or some incremental change in the depth of the sound stage, or some jitter artifact, and sometimes I think I hear something, but then I always start listening to the music and lose concentration. It happens almost every time.

Tim

 

[Ethan Winer]

At the time, our knowledge of jitter was not robust and we had no idea that the human ear could hear picosecond differences.

I'd love to see some proof of that. Jitter manifests as sidebands, which is basically noise that's typically 100+ dB below the music. There are so many other reasons people might believe they hear a change that doesn't exist, I'll go with Occam until shown evidence to the contrary.

Many suggest the Nyquist Theorem is absolute proof that 16/44.1 can capture everything to 20khz. In my experience that is simply not true.

Are you really disputing Nyquist? Some circuit designs may come up short in their implementation, but you'll have a very difficult time proving Nyquist (and Shannon and Fourier) wrong!

--Ethan

 

[Ethan Winer]

I haven't made any comment on null tests in this thread, Ethan.

But that's the subject of this thread. That, and what specific physical properties of sound digital recording cannot capture.

But these are both software solutions. Their performance will therefore depend on the hardware with which they are used.

Of course, but they're still highly useful tools. And they can easily settle the question of, for example, at what level are artifacts no longer audible? If I could afford an AP analyzer, believe me I'd own one!

--Ethan

 

[Ethan Winer]

I like Ethan's inquisitiveness but I don't think he will ever be satisfied with "proof" of what I hear.

Right, it's impossible to know what someone else hears. Or, more accurately, what someone else perceives. But:

The hirez cello sounds like the real thing. The redbook not so much. It is missing the "hall" effects. It is missing the woodiness of the cello. The timbre is simply off.

I'd love to see the residual from a null test, which should be easy to produce. You have a high-res file that you down-res'd to Redbook, yes? So the two files should track exactly with no drift, making a null test easy and conclusive. Can you load both file into a DAW and export the difference? I'd love to see a snippet as a Wave file. "Hall effects" would be a function of the noise floor, and timbre is clearly frequency response. So assuming the down-res was done correctly, any differences should show up clearly.

--Ethan

 

[microstrip]

(...) But some of the stuff you guys talk about, I just don't have the patience to hear. (...)

Tim

Perhaps because your system is assembled in a such way that you need concentration to hear it. Or, just beacause you valuate some other things so much that it is difficult for you to concentrate on this "stuff".

BTW- audiophiles also are of the type "what works for me". If you look at the systems we have in this forum you will find that they differ a lot. But as you say some of us have "some stuff" in common.

 

[Ethan Winer]

Yes, Ethan... now you are being dense ... if you're "honestly don't understand what I'm showing", then I think you should seek another line of work.

There's no need for an insulting tone Bruce. I'm not the only one here who finds your writing less than clear, for example when you initially described your idea for a null test. No matter what else you may think about me, I am never evasive. I'm not sure how busy you are, but I don't have the time to dig around for meaning among 29 pages of forum posts, or try to guess what an unexplained equipment list means. Also, I'm not familiar with the type of FFT you showed. I'm used to seeing data expressed as a graph of dB levels versus frequency, as opposed to trying to decode those numbers from color gradations.

Bruce, I just went through this entire thread twice, and I still don't see you state what brand / model players you measured or what / how you measured. I see the list on the Boston Audio site of the three players M&M used. But I can't tell what specific frequencies are "missing" from the players in your graphs, or how much attenuation there is versus freq. I also don't know how many of the M&M tests used which players. As I mentioned earlier, even if 3/4 of the M&M tests were not valid, that still leaves an awful lot of tests using players (I assume) you approve of that still showed nobody could hear a difference. Further, if the three players M&M used are typical popular models that many people own, why isn't that a valid part of the comparison? This is no different than testing the audibility of ultrasonic content using normal speakers that can't output beyond 20 KHz, which is most speakers! Yet people claim to hear a difference anyway.

--Ethan

 

[Ethan Winer]

I just looked at the measurements published by Paul Miller at HFNRR and John Atkinson at Stereophile.

Without a link I have no idea what measurements you're referring to. And that doesn't answer the question of how you personally concluded that artifacts 90 dB below the music are audible.

I am saying that no difference was seen in the distortion measurements.

So how do you know they sound different? Your "answers" so far raise more questions than they settle.

Curious, my measuring system was listed in the post.

What post? I just traced back through all of our exchanges, and I saw nothing more than your eMu sound card mentioned. So again I ask you, please describe in detail the specific experiments you performed to conclude that you can hear stuff more than 90 dB below the music while the music plays.

--Ethan

 

[Bruce B]

Bruce, I just went through this entire thread twice, and I still don't see you state what brand / model players you measured or what / how you measured. I see the list on the Boston Audio site of the three players M&M used. But I can't tell what specific frequencies are "missing" from the players in your graphs, or how much attenuation there is versus freq.--Ethan

Let me be more clear then...

1. I used the same exact program you claim to use ... Sound Forge. These graphs are from spectral analysis/FFT plugin. Guess you either need to RTFM or stop claiming you use Sound Forge when you obviously don't know how to read the graphs or know what I'm talking about. Spectral analysis can give you lots more information than a normal FFT graph.
The freq is on the left, the timeline goes across and the amplitude is in colors!!!

2. I've stated twice or 3 times now that I used the same exact brand/model SACD players they used in the tests and also stated I recorded the audio output stream.

3. I have provided proof that this test was flawed. You have showed me nothing... nor any of the other members. You always ask for proof and when someone lays it out before you , you claim you're too busy to follow through.

4. Tell us more about acoustics, and stay out of electronic and null discussions where you are truely at a disadvantage.

5. Where is that ignore button?

 

[andy_c]

I'd love to see the residual from a null test, which should be easy to produce. You have a high-res file that you down-res'd to Redbook, yes? So the two files should track exactly with no drift, making a null test easy and conclusive.

I've done some similar experiments using Audio DiffMaker. My sound card requires me to go into the sound card setup software when the sample rate changes from one file to the next, making mixed sample rate music in the same playlist impractical. I had some 24/96 DVD-A rips and wanted to see what might be lost if they were downconverted to 16/44.1.

The idea was to browse some DVD-A rips and find a representative song that had more high-frequency content than the rest. Then, the song would be downsampled to 16/44.1, then upsampled back again to 24/96. This "round trip resampled" file would be compared to the original and the difference taken using DiffMaker.

I first tried this with the freeware version of the r8brain resampling software. I found the difference file had annoying clicks in it, indicating the resampling might not be transparent. So I looked for more powerful resampling software. I found SoX. When I first tried SoX, I did downsampling with triangular PDF dither and a scale factor of 1, using the highest-quality resampling algorithm offered. It immediately told me there were clipped samples in the resulting file. SoX allows scale factor adjustment, so I made the scale factor as close as possible to 1 while avoiding clipped samples. Let's say that scale factor were 0.945. This was adjusted so that if I had used, say, 0.946, there would be one or more clipped samples in the file. Once I got the scale factor right, I upsampled again to 24/96 and used DiffMaker to compute the difference between this file and the original 24/96 DVD-A rip. DiffMaker figures out the necessary time shift and scale factor to optimize the null (including sub-sample time shifts), and it did quite well, computing the exact scale factor I had used with SoX to prevent clipping. When I listened to the difference file with my headphones, I couldn't hear anything at all, not even noise. Not enough gain I guess. So I played it through my speakers with the volume all the way up and my ear right next to the tweeter. I could then hear a slight increase in noise above the analog thermal noise of the electronics, but nothing whatever that sounded like music or distortion.

I don't consider this any kind of comprehensive proof of transparency of downsampling to 16/44.1 though. For example, I've heard that people who make ultra-high dynamic range recordings of fireworks (I think Mark "Basspig" Weiss and Tom Danley have done this), and found the dynamic range of the 16-bit version to be wanting. At the same time though, my experience with this makes me rather skeptical of claims of obvious sonic degradation with music.

 

[amirm]

Guys, let's take the emotions down a notch . It is just a conversation and should not raise anyone's angst.

 

[Phelonious Ponk]

Perhaps because your system is assembled in a such way that you need concentration to hear it. Or, just beacause you valuate some other things so much that it is difficult for you to concentrate on this "stuff".

My system was assembled by a very talented engineer who I'm certain did a better job than I ever could. I vote for the valuation thing.

Tim

 

[fas42]

So I played it through my speakers with the volume all the way up and my ear right next to the tweeter.

Hmmm ...., where have I heard of that technique before ...

When I listened to the difference file with my headphones, I couldn't hear anything at all, not even noise.

Of course, that tells you absolutely nothing. You now have to do it thousands of times, with thousands of different recordings, listened to by thousands of different people, in thousands of different test environments, to even have a vague chance of starting to mean anything possibly significant to a rational person ...

Frank