Cat Stevens - Tea for the Tillerman HDTracks

[bottlerocket]

 

 

Click to enlarge

 

 

 

 

Click to enlarge

 

 

[Matias]

Great, so this one looks real. How does it sound?

 

[DaveMan]

First listen bears out the dynamic range shown in the screenshot above. This reminds me of the uncompressed version of Band on the Run. It's almost unnatural; we're so used to listening to compressed recordings (especially familiar ones like this and BotR!). What a refreshing change! I'd buy it again.

 

Good job Ted Jensen and HD Tracks!

 

[DJSY]

foobar2000 1.1.10 / Dynamic Range Meter 1.1.1

log date: 2012-01-19 12:40:24

 

--------------------------------------------------------------------------------

Analyzed: Cat Stevens / Tea For The Tillerman

--------------------------------------------------------------------------------

 

DR Peak RMS Duration Track

--------------------------------------------------------------------------------

DR14 -0.01 dB -19.23 dB 3:52 01-Where Do The Children Play?

DR13 0.00 dB -18.11 dB 3:48 02-Hard Headed Woman

DR12 0.00 dB -14.90 dB 3:20 03-Wild World

DR12 -5.10 dB -20.89 dB 3:46 04-Sad Lisa

DR12 0.00 dB -15.39 dB 3:38 05-Miles From Nowhere

DR12 0.00 dB -16.93 dB 1:59 06-But I Might Die Tonight

DR13 0.00 dB -16.75 dB 3:06 07-Longer Boats

DR13 -2.71 dB -18.67 dB 3:25 08-Into White

DR14 0.00 dB -17.86 dB 5:08 09-On The Road To Find Out

DR14 0.00 dB -18.83 dB 3:41 10-Father And Son

DR15 -1.15 dB -23.08 dB 1:03 11-Tea For The Tillerman

--------------------------------------------------------------------------------

 

Number of tracks: 11

Official DR value: DR13

 

Samplerate: 192000 Hz

Channels: 2

Bits per sample: 24

Bitrate: 4272 kbps

Codec: FLAC

================================================================================

 

[Matias]

But still, comparing subjectively the MFSL and the HDTracks, what is your opinion?

 

[esldude]

I don't yet have the HDTracks. The posted info on them for this recording mean I probably will get them as they do appear to be genuine hirez material from master tapes. I'll try to report back at some point.

 

This recording was of good quality to begin with though not best ever. I wonder about the gold Ultradisc having no response above about 17.5 khz. MFSL said they used the direct mastertapes without compression or EQ directly into a special AD converter. I believe in those days they also attempted when possible to get identical tape machines used in the recording to play those master tapes. Will be interesting to compare in any case.

 

[goldsdad]

"I wonder about the gold Ultradisc having no response above about 17.5 khz."

 

In the attached screenshot, which track best resembles your Ultradisc rip?

 

Click to enlarge

 

 

 

[esldude]

Mine looks more like the upper one. Just realized looking at your and my previous screenshots I had the upper limit displayed set for 8khz(why that low I don't know probably some other project).

 

The idea there is little above 17.5 khz comes from FFT analysis of the track. I will redo the screenshots and post a version more useful for comparison here in a couple minutes.

 

[esldude]

Fixed the plots so you can see it for better comparison with what you have.

 

This as I said, was from a gold MFSL UltraDisc II CD.

 

As well as the spectrum view I have included the Plot spectrum view. My cursor is near the top of what is shown in the histogram of frequencies. 17,641 hz.

 

 

BTW is your plot from the regular version of the CD?

 

 

 

 

 

 

[esldude]

Just for some comparison I have the same info from MFSL Gold CD of the same recording.

 

The hi-rez dynamic range looks much larger in the linear scale, but really is maybe only 3dB more in the hi-rez version. I am including the same using a dB scale as well.

 

Another inclusion is using the envelope tool which shows the max dynamic range used for a track.

 

The FFT of this recording shows a max frequency of about 17,500 hz, so the hi-rez version certainly has more upper frequencies as it should.

 

Thanks for posting this as it may prompt me to buy the HD version. The MFSL is quite a good recording itself of course.

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

[wgscott]

It looks like it goes to 22kHz to me. I think the "frequency analysis" cross-section plot isn't going low enough to show some of the signal that is clearly present in the fft.

 

[goldsdad]

My spectrograms are of the Ultradisc, too. The top track is the lossless rip and the bottom track is a conversion to lossy AAC. My worry was that you had accidentally created lossy files when ripping or at some later time because you said there was nothing above about 17.5 kHz.

 

A spectrum plot can be misleading because it is an average of the selected region (or some maximum number of samples, whichever is the least), so frequencies which exist for duration much less than the total duration of the region can be averaged almost out of existence and become lost in the noise.

 

A spectrogram is like a sequence of many very brief spectrums standing alongside each other. This normally reveals the highest frequency in the signal distinct from the noise.

 

 

[esldude]

Yeah, I think goldsdad and wgscott are right.

 

I often zoom in on the waveform view, but not usually on the spectrogram. Doing so does show when zoomed in there is very, very little above a line around 17.5 khz. So as you say probably just too little to show up in the frequency analysis view. I think not zoomed it probably makes it look like there is more high frequency info than there really is though there definitely is some.

 

Since you ripped from the same disc goldsdad here is the MD5 hash for that file. It was ripped as a wav from the CD. Any metadata may cause the hash to differ even if you change it to a .wav, but in any case might be interesting to see if they match.

 

ee88a407694c3f5e94dcc139a8387db0

 

 

 

[goldsdad]

"...when zoomed in there is very, very little above a line around 17.5 khz."

 

Depends on where you zoom in. Some intervals, such as the first 2 seconds have almost no signal above 17.5 kHz. Zoom in to a 2 seconds interval at about 3 minutes 53 seconds to see plenty signal up to 20 kHz.

 

Click to enlarge

 

 

 

Click to enlarge

 

 

 

 

Below is the spectrogram that you posted and to which I've added a black dashed line at the cutoff at about 20 kHz in case anyone reading this is unsure of what to look at.

 

Click to enlarge

 

 

 

 

My rip is verified with the AccurateRip database but you are correct that the md5 is different to yours which may also be verified. By the way, XLD usefully writes WAV without embedded timestamp or tags.

 

 

 

 

 

[esldude]

Okay, I downloaded the HDtracks 192 khz recording for Cat Stevens yesterday.

 

Did some comparing in Audacity this afternoon. Yes, first I listened to it extensively as well as the MFSL version. Plan on some other people listening and giving me their opinion before I say what I am hearing.

 

Had to cheat a bit for a good comparison. I down-sampled the 192 version to 96. I up-sampled my MFSL CD to 96. The 192 version was louder as well. On average around 4 dB. So I then normalized both for the 3rd track, Wild World. I then low pass filtered the 192 version at 20 khz. Put the left channel of the filtered, normalized, downsampled 192 and the 44.1, upsampled left channel in the same window as stereo tracks. Lined them up carefully based upon some early sharp spikes in the waveform.

 

I am showing small versions of the screenshots here, click below for the large version.

 

First up is the log spectrum view. They are pretty darn close.

 

 

 

Next up is the log waveform view (easier to compare in log form). Again quite close.

 

 

 

 

Then we have a very zoomed in view just as the music starts near the beginning. Those little diamonds are individual bits you are looking at there. Though not an exact match this is impressively close considering the differences are hugely magnified here and we are talking about a handful of samples varying by a handful of bit levels different.

 

 

 

 

Finally we have the same zoomed in view near the end of the song in a highly energetic area. I had to trim things up on that end. The speed of these tracks is also slightly different. Again not 100 % exact, but considering all the processing to get them together this is very very close.

 

 

 

 

Though not shown I did something of the reverse. Taking the original 192 khz 3rd track, I high pass filtered it at 20 khz. Basically seeing what is left. There are bits up there for sure. Plenty of information. I listened to that high pass filtered file. I could hear nothing, absolutely nothing even with headphones. I am over 50 and my hearing might make it to 15 khz or just short of it even. Still makes you wonder what the higher rates are getting you if anything. If I couldn't hear anything with gain cranked up over headphones and the fundamentals gone so they aren't masking it, is there any chance any of this would possibly matter to me upon normal playback?

 

[esldude]

 

Okay, my report on what the 192 khz version sounded like to me versus the MFSL Ultradisc II gold CD.

 

From my previous posts I would imagine most were thinking I would say I heard no difference. Especially after my concluding paragraph in the previous post. I filtered out everything below 20 khz, and what was left is completely inaudible to me. Further filtering out everything above 20 khz and comparing to redbook CD came far closer to matching even down to the bit level than I would ever guess.

 

Yet, I thought the 192 khz download from HDtracks was clearly a nicer recording. How do you explain that? Hmmm, not sure I can. Not certain it isn't just fooling myself....the old placebo effect.

 

Couple of quotes:

 

The first principle is that you must not fool yourself and you are the easiest person to fool.

Richard P. Feynman

 

 

For a successful technology, reality must take precedence over public relations, for Nature cannot be fooled.

Richard P. Feynman

 

Yet, I thought the 192 was simply better. Not a tremendous mind blowing difference, but clear and without much doubt. A couple of friends listening over a high resolution system thought the same thing. Played the MFSL CD first, then the 192 version. In less than 15 seconds one fellow said, "that sure sounds better, it sounds like it has more information. Everything sounds real." He has no experience with computer audio and didn't really know what the recording formats were. The other friend immediately agreed with similar comments. I felt the same when I first listened, but had told none of this to them.

 

Now we did not listen to my filtered 192 khz version where all response above 20 khz was filtered out. We did listen to upsampled files from the CD, and they seemed no better. I do wonder if the high sample rate with all the high frequency additional info removed will still sound better, or the same or what.

 

Leaves one with the same conundrum all audiophiles have subjectively. Could any of us pass the DBT on this? Results of other tests make me think we shouldn't expect to. Yet it seems so simply clear one sounded better than the other you have this gut feeling there is something to it. If you then fail the DBT, should you believe it was placebo, or do you trust your gut and insist the DBT must be wrong, or inadequately revealing, or that it corrupts the normal listening process?

 

There is one other alternative. That is that the high sample rate recording was remastered better. The test for that is to filter ultrasonics and downsample to 44.1 khz then see if it still sounds better. I don't know for certain, but MFSL claimed to do no EQ, no compression, nothing other than transferring the master tape straight to CD without anything being done. I suspect the 192 version was done the same. All interesting of course.

 

And once again shows just how hard it is to trust your knowledge while doubting your gut sensations.

 

 

 

[Julf]

I know it is a rather silly and basic thing, but is there any way the loudness (either absolute volume or degree of compression) could have been different?

 

[wgscott]

If you filter out everything below 20kHz, why would you have expected to hear anything at all? My hearing goes to 16kHz, max, and for my age, that is reasonably acute.

 

If you removed everything above 22kHz, the two files are presumably still different in that the high def one is 24 bit, not 16 bit. This is likely where most of the audible differences between the two files would be -- i.e, within the dynamic range.

 

As noted above, the loudness or whatever it is called would need to be normalized for a fair comparison, and you would need to know whether the track was remastered.

 

I think the higher-than-22kHz stuff is important primarily in the sense that you want to move aliasing artifacts out of the audible range.

 

[esldude]

There was indeed a loudness difference. About 3.8 dB louder on the 192 recording. The system we were listening over at my friend's house controls volume digitally to .1 dB. I had made sure of how much difference there was in the first couple tracks we auditioned. So I had the remote and changed volume appropriately when we changed tracks. Might not have been an exact match, but wasn't a big general volume difference.

 

Now when I first compared them at home the newer 192 sounded much better of course. And in just a few seconds of playing them I knew there was an obvious volume difference. So I made those same adjustments comparing them at my home after the listening to the first track.

 

Compression didn't appear to be obviously different looking in audacity. MFSL claimed to use none, the other I don't know, but it doesn't appear that they did. Another situation where knowing the provenance of a hi-rez recording would be nice.

 

[esldude]

I didn't expect to hear anything with everything below 20 khz filtered out. It just struck me to do it and see if any artifacts were left in the top of what I could hear. I remember in early CD days listening over head phones when my hearing still went to around 17 khz. I could play most any test tone at moderate level between 15 and 20 khz. The filtering was so poor you would hear the difference tone fold back down well into the audible area. So just wondered if anything like that was going on though I doubted it.

 

Yes, one was still 24 bit and the other 16. As they were sourced from reel to reel I doubted that would matter actually. MFSL claimed no compression or EQ, it looks likely however the 192 khz was done that no compression or expansion was used either. I don't know that. My thinking is either version has dynamic range well beyond the master tape.

 

The higher bandwidth would of course move aliasing products further out of range. But we also listened to upsampled CD tracks both at 88.2 and 96 khz. Really heard no difference in either. Any aliasing effects with the 22 khz and below range at the AD end of course would still be there.

 

There was a definite speed difference. It isn't huge in absolute terms, but the 192 version is a touch slower. I also noticed when zooming in on the files in audacity there were other things. I am thinking we are talking about flutter from the tape machines. Periodically there were extra squiggles in the MFSL version vs the 192. I didn't bother to check on the periodicity of it. But it would come for a couple of milliseconds and then go away for much longer than that. In between those you got waveforms like those I posted above that are surprisingly close to identical.

 

Of course this is a little bit of an apples and orange situation. My most likely guess is we are just fooling ourselves somehow. Even if we could prove we are hearing something real, well then the files most likely aren't the same in ways that have nothing to do with bandwidth or bit depth. I am actually pretty impressed they are so darn close even at a micro-level considering they were done at different times by different people on different equipment. Or so I assume. Any chance HDtracks obtained the original MFSL digital masters and put them in at the higher sample rate? Don't know the answer to that.

 

 

 

[goldsdad]

"filtering out everything above 20 khz and comparing to redbook CD came far closer to matching even down to the bit level than I would ever guess."

 

Yes, far too close a match in one image. I'll come back to that below.

 

 

"Yet, I thought the 192 was simply better. Not a tremendous mind blowing difference, but clear and without much doubt. A couple of friends listening over a high resolution system thought the same thing."

 

I'm not at all surprised that the MFSL (16/44) and HDTracks (24/192) sound different (regardless of resolution). Each was transferred from analogue tape to digital and mastered by different engineers using different techniques and equipment. Surely, it would be incredible if they did not sound different.

 

 

"once again shows just how hard it is to trust your knowledge while doubting your gut sensations."

 

In this case, safely trust your gut, I think. My suspicion is that there has been a mistake in your production of the close-up comparison of waveforms near the end of the left channel of each version of track 3, i.e. the upsampled MFSL and downsampled low-pass filtered HDTracks.

 

 

Here's the suspect image again.

 

Click to enlarge

 

 

 

 

Supposedly, one channel is the result of your downsampling and low-pass filtering a transfer and mastering that was produced by one engineer using his techniques and chain of equipment. And the other channel is a result of your upsampling a transfer and mastering that was produced by another engineer using other techniques and equipment.

 

As if by a miracle, the channels appear to precisely match except for being misaligned by one sample!

 

This image shows the result of overlaying the channels.

 

Click to enlarge

 

 

 

 

Then shifting a channel by one sample gives an overlay which is perfect except for quantisation errors in the pixelating of the original graphic.

 

Click to enlarge

 

 

 

 

Doesn't that seem like an incredibly improbable coincidence? Too good to be true? Could there have been a mistake in your procedure? Bear with me for a moment more.

 

To emphasise the improbability of that occurance, here are before and after screenshots of a little test. The four tracks are all the left channel of the MFSL track 3. The before image shows the tracks which differ only in that they have had 0, 1, 2 and 3 samples cut from the start of the track and then been aligned. The after image shows the result of upsampling the tracks. The upsampled waveforms are extremely similar to each other but close inspection reveals the corresponding samples to be in significantly different locations. (The fourth track matches your result for that region in an earlier post.)

 

Click to enlarge

 

 

 

Click to enlarge

 

 

 

So, if my upsampling of tracks which differ by just one sample can produce waveforms with clearly different samples, how can it be correct that your upsampling of one mastering and downsampling and filtering of another mastering can produce identical waveforms except for a misalignment by one sample?

 

Furthermore, your comparison of the waveforms over the interval from 40 seconds to 2 minutes shows differences as expected. And your close-up comparison near the beginning (same region as in my own test) shows distinct differences as expected. That adds to the improbability of your comparison near the end of the track revealing identical waveforms.

 

I hope you don't mind very carefully rerunning your comparison of the suspect region.

 

 

 

 

 

 

[esldude]

Yeah, could have mistakenly put the same track on both without meaning to do so. I have wondered if that were the case. Will probably get time tomorrow to do it again. I may even have saved it as an .aup file. So I can compare the one posted with another run at it.

 

[esldude]

Good catch Goldsdad

 

A good example of peer review I suppose.

 

I believe all the images I posted were correct other than the one near the end of the third track. It is just too close a match. I was checking a number of things and must have pasted the same track in both positions somewhere in the middle.

 

I did the same thing from scratch this time. I resampled both sources of the third track to 96 khz. I filtered the 192 khz track by removing everything above 20 khz.

 

I am attaching an image from near the beginning and three more near the end. I will also put one large image from near the beginning in the links below. I am still impressed with how close the waveforms are at this highly zoomed in level. Each image is looking at a few dozen bits when we are getting well over one million per second. Because of trimming for the 192 track being slower than the MFSL tracks I wasn't able to capture the exact same position as the previous image. But it is in the same area.

 

This from near the beginning. Very close waveform shapes, but not identical and bits are different.

 

 

 

This from near the end of the track. Again impressively the same considering it was mastered years apart. But not identical.

 

 

 

Yet another near the end of track 3.

 

 

 

One more which is impressively the same though not identical.

 

BTW, my apologies for messing this up the first time. Kudos to Goldsdad for seeing it and pointing it out. Thanks to Goldsdad as well.

 

 

 

 

 

[DJSY]

Here is a High Rez Spectrum of the HD-Tracks 24/192K file of On The Road To Find Out. What do you think the spikes around 60Khz are? Tape Bias?

 

 

[esldude]

Tape bias I believe was 10 khz for 15 ips and proportional to tape speed. So I doubt it is bias. It also is very low in level like around -90 to -100 dB. It also shows up in the other tracks. So not sure what it is. Could be any number of things.