Why Does Some High Rez Sound So Bad?

[Flecko]

The problem with remasters is that the sound is compressed, that could be received as better.

And there are a lot people which like the compressed sound or simply don't care or hear it. Music is not meant to be  played on real hifi systems anymore. It is meant to be played on electret speakers of a mobile, while driving in the bus or in the train. It is great for those guys if it is just loud. I met somebody who asked me, if it would sound better, if he would buy a cd instead of listening to download mp3 with his earphones and his mobile. What should I say??? That is how the mainstreem listens to music...

[pedal]

Pedal, try the 16/44 vs 24/96 version of Joao Gilberto and Stan Getz, just curious about your findings

Finally I had the chance to play around a little with my software. 

I bought the original CD when it was issued middle of 80s. (In fact it was Verve who woke up the jazz lover in me. I bought practically all the classic Verve reissues on CD at that time, falling in love with many of the jazz giants like Oscar Peterson, Stan Getz, Billie Holiday, Louis Armstrong, etc). I know these albums quite well.

The 2 frequency plots show that the CD has the mandatory lowpass filtering at 21kHz.
The 24/94 download from HD Tracks has twice the bandwidth, extending to about 44kHz, indicating this is the same master as developed for the SACD issue.

Also, I performed a TT DR analyze of both. I have the licensed software for Mac, usable also for 24/96 material.
-There was a slight difference between the 2 versions, but both got DR=11, which is a fairly good result for this kind of material. Looking at the spectrum, there are no signs of compression.

Conclusion: The 24/96 download of Getz/Gilberto is the real deal. And it sounds very good (for what it is; a 1963 recording of a small jazz ensemble).


(PS: The frequency analyzes are not direct comparable because they are just printscreens with different scaleing/timewindow).

[PeterSt]

Hey Pedal. Great.

Please allow me to come up with a few very general remarks, so you may be able to interpret all better a next time (and note I may be saying things you knew all along).

I was not behind the knobs here, but looking at two of these "static" pictures I would wonder why (e.g.) the 13KHz mark from the first picture is at around -45dB while in the second this is at -62dB or so.

The second picture has it's lowest output around at around 20KHz, and then rises ?? No ... This means that it is noise already. This by itself would tell me that no real frequency output is there above 20KHz, and since it is not 22050Hz but under 20000Hz, it would be my suggestion that this hires has been low passed very nicely ...

The 2 frequency plots show that the CD has the mandatory lowpass filtering at 21kHz.

Haha, yes, that is what one would say. But no, a CD just has NOT any low pass filtering, because there is nothing to low pass. It's just not there above 22050Hz. That is, when Anti Imaging has been applied by the "analyser". However, looking at the left halve of the CD plot, there's harmonic distortion visible, which tells me that no AI filtering has been applied (again, by the analyser). *This* now tells that indeed low pass filtering must have been there (analyser !!) because otherwise the shown frequencies would have continued after 22050.
All 'n all a bit tricky to use this ...
(disclaimer : the HD visible - very theoretically can be the overtones of the instrument concerned when this was a too small snapshot. This is why it needs to be behind the knobs really)


So ... without being behind the knobs, it would be my conclusion that this is exactly not the real deal. And worse, where the leftmost part of the plots have the same output level, the more going to the right, the less output the hires has. You know what this means ... filtering (and not even steep).

Maybe I am wrong all over, but I don't think so. Maybe it is better to use a Waterfall plot.

[achri-d]

So ... without being behind the knobs, it would be my conclusion that this is exactly not the real deal. And worse, where the leftmost part of the plots have the same output level, the more going to the right, the less output the hires has. You know what this means ... filtering (and not even steep).

Maybe I am wrong all over, but I don't think so. Maybe it is better to use a Waterfall plot.

Yes filtering - reconstruction and noiseshaping.

The two following pictures show a spectral content likely to be as it should (24/96):



Rgds.

[PeterSt]

That SURE looks more like it !
(and nice you used the same program)

Thanks.

[pedal]

Maybe I am wrong all over...

Yes, you are wrong all over.

Like I said in the PS: The 2 plots are not comparable, because:
-The Spectrum Analyzer feature of (my version at least) Audacity has a limited capacity. It can spectrum analyze only 24 seconds of a CD (16/44) and about 10 seconds of a 24/96 file. What you see in my previous post is the beginning of track #1 The Girl From Ipanema.
So, you have an average of the 20 first seconds of the CD, and an average of the first 10 seconds of the Hirez. So it becomes apples vs oranges.

Further more, if you listen to the start of this song, there is only a softly male voice and a low key acoustic guitar. In other words, no high frequency content anyway. On top of it, its recorded in 1963 with limited bandwidth microphones, probably rolling off at 15kHz. The recorder/mastertape in use probably didnt capture much content above 18kHz eighter.

The spike above 14kHz is interesting. Analyzing samples from different places throughout the song, it seems it is present only with vocal (not during sax playing), indicating that the voice microphones they used had a severe high frequency resonance.

On an old recoridng like this there is very little high frequency content. What you see above 15-20kHz is mainly noise. The particular noise pattern you see on the 24/96 is probably the noise shaping of DSD mastering. I have seen it on others too. The bandwidth limit of 44 (half of 88.2 - often seens in SACDs) confirms this indication.

---------

Repeated A/B* listening now to The Girl From Ipanema in CD (ARCx8) vs 24/96 (ARCx4) I find the differences quite small. But the 24/96 is slightly better. You can hear Gilberto's voice a little clearer, you get a better sense she is standing in a separate vocal boot than the rst of the band. Also the cymbal in the right channel, when the sax starts soloing, is more distinct.

*I hate A/B comparisons. After a few times repeated listen to the same short track they all start sounding the same. 

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

Summary, there is nothing "wrong" with this 24/96. It is ucompressed. No added EQ. It sounds marginally better than the CD, although the age and recording doesnt offer much high frequency anyway. Finally the price is very symphatic, costing $18 which is less than the CD in the shop.

[pedal]

Hey, we got company from my friend achri-d, a very experienced and technically educated hi-fi enthusiast. That's nice! I hope you can contribute more to this forum, Christian, since you have jumped on the high-rez band wagon too!

--------

GENERALLY SPEAKING: Although 50 years old recordings doesn't contain much high frequencies, I still find the 24/96 rewarding because (when done right) you avoid the bogus downsamling done to CD.

Normally the old mastertape is transfered to 24/96 (or 24/192). Then they do some (more or less) editing in the 24/96 domain using programs like ProTools etc). Then it is noiseshaped and downconverted to 16/44. This process is rather harming to the SQ. Releasing it straight as a 24/96 file is better, even if the recording is old.

I had an interesting correspondance with British Producer Tony Faulkner, and he explained me these things quite detailed. He was very clear about most downsamplings to 16/44 was badly done, using "bogus" software. He was also sceptical to noise shapping. He also said that if recorded in 44kHz, a CD could sound very good. I think Faulkner is one of the most experienced producers in this field. You have probably read interviews with him in Hi-Fi News and Stereophile.


@achri-d: The last graph you posted, did you do it with the Adobe Audition program?

[GerardA]

Very interesting, Pedal!

But does this mean that old recordings which have no output above 20 kHz don't need any higher samplingrate than 44 kHz?
This is not my experience/idea. If so, how to explain that?
I'm sure the 384 kHz of the NOS1 is doing more than only counteracting bad quality software of the recorders...

[achri-d]

I had an interesting correspondance with British Producer Tony Faulkner, and he explained me these things quite detailed. He was very clear about most downsamplings to 16/44 was badly done, using "bogus" software. He was also sceptical to noise shapping. He also said that if recorded in 44kHz, a CD could sound very good. I think Faulkner is one of the most experienced producers in this field. You have probably read interviews with him in Hi-Fi News and Stereophile.

@achri-d: The last graph you posted, did you do it with the Adobe Audition program?

Hello Petter.

I have some CDs engineered by Faulkner and some of them on the Hyperion label are very good indeed - for instance: Bantock: Hebridean Symphony, Celtic Symphony, The Witch of Atlas, The Sea Reivers - Vernon Handley, Royal Philharmonic Orchestra. Hyperion CDA66450. 1 CD. 1991.

I use the latest version of audacity to create the second graph. I have developed a small C#/Matlab application in order to verify the Audacity -and  it works as expected.

Rgds.

[PeterSt]

I have developed a small C#/Matlab application in order to verify the Audacity -and  it works as expected.

But don't give it FLACs for input !!

[achri-d]

But don't give it FLACs for input !!

Hello, I always do analysis on wavs- these are simple to handle (known by Matlab for instance) and audacity treats them with ease. Rgds

[PeterSt]

But Pedal, if it is all like you described in your last post, all is moot ! I mean, how does this proove that this particuar album is OK ?
I don't see it, and instead see the contrary. So, your explanation may be ok, but now it says nothing.

Don't you agree ?

[PeterSt]

Hello, I always do analysis on wavs- these are simple to handle (known by Matlab for instance) and audacity treats them with ease.

Ok. I only said it because with FLACs AudaCity fails, and won't show content above 22050 ...

[pedal]

But Pedal, if it is all like you described in your last post, all is moot ! I mean, how does this proove that this particuar album is OK ?
I don't see it, and instead see the contrary. So, your explanation may be ok, but now it says nothing.

Don't you agree ?

Of course, when the source it self doesnt contain (almost) any music signal above 20kHz it's hard/impossible to meassure any difference. I have also the Adobe Auditon program. There I can zoom down to the very 1/96000th sample of the sound file. I did it to study square waves on (some modern!) titles I have.

If I find a passage on the Getz/Gilberto album with a clean hit on the cymbals or something, then I might can zoom in on that very spot, precisely the same place both on the CD and 24/96 file to check for differences in the high frequency response. ...But when going back/forward to Vista OS last week I lost all my installation, so it has to be redone some day I have time to find the program disc. Right now I rather give priority listening to my NOS1...

[PeterSt]

*I hate A/B comparisons. After a few times repeated listen to the same short track they all start sounding the same.

See ? we agree all the time !