Artefacts in hi-res audio?

[Boye]

Being curious as to why some people claim that hi-res audio files "sound better" than their redbook counterparts I tried to compare them using Audacity (as it is free to use).
Some files were clearly mastered louder (seen on the waveform) and with different eq (seen on the spectral analysis). This was particularly a phenomenon on SHM-CD versions of some records.
But other files looked quite similar overall to the CD versions. However, some artefacts seem to appear in the supersonic range of some tracks. The attached images shows a comparison between two similar tracks, one being a 24/192 version that has some strange information in the 30 kHz to 80 kHz range when comparing the two spectra.
Could this be a result of the file settings in Audacity or coud it be some questionable information stored in the file?
Mind you, this is just an example. I have seen this phenomenon on several files, so I began to suspect that my analysis software was to blame.

 

[Vincent Kars]

I always inspect hires as often there is a lot of garbage in it.

A true 192 kHz but some weird spike and a rising noise due to the noise shaping.

A bit more: https://www.thewelltemperedcomputer.com/SW/AudioTools/Spectrum.htm

 

[Boye]

I'm probably struggling to stay afloat in the deep end regarding my knowledge and understanding of digital audio. When I fiddle with the 'size' setting in the Plot Spectrum the artifacts seem to disappear, so perhaps I am drawing conclusion on the wrong basis. Notice that I use the logarithmic scale.
I probably have to look a little more at your shots.

 

[SSS]

Being curious as to why some people claim that hi-res audio files "sound better" than their redbook counterparts I tried to compare them using Audacity (as it is free to use).
Some files were clearly mastered louder (seen on the waveform) and with different eq (seen on the spectral analysis). This was particularly a phenomenon on SHM-CD versions of some records.
But other files looked quite similar overall to the CD versions. However, some artefacts seem to appear in the supersonic range of some tracks. The attached images shows a comparison between two similar tracks, one being a 24/192 version that has some strange information in the 30 kHz to 80 kHz range when comparing the two spectra.
Could this be a result of the file settings in Audacity or coud it be some questionable information stored in the file?
Mind you, this is just an example. I have seen this phenomenon on several files, so I began to suspect that my analysis software was to blame.

Some time ago I viewed the output of CD-res and hi-res files and PCM-converted DSD. And it is true that some hi-res files and the PCM-converted DSD contain noise above 20 kHz. Whether the extra noise influences the sound due to effects in the amplifier electronics or not I cannot say. If an amplifier has an input filter slightly above 20 kHz then the noise may not have an effect.

 

[Boye]

True, but I can't help but wonder where it comes from, especially in the form it appears in my screenhots - or if they even convey the truth. I mean, if the files are sourced from the original master tapes, there might be some signal and noise in the supersonic range, but I doubt it would show like a hump in the supersonic range.

 

[pkane]

True, but I can't help but wonder where it comes from, especially in the form it appears in my screenhots - or if they even convey the truth. I mean, if the files are sourced from the original master tapes, there might be some signal and noise in the supersonic range, but I doubt it would show like a hump in the supersonic range.

That is often due to noise shaping and/or an imperfect method or implementation of lowres to hires conversion.

I posted this example recently -- this is an extreme case of noise shaping that occurs with DSD content:

 

[Geert]

Could this be a result of the file settings in Audacity or coud it be some questionable information stored in the file?

Amir did a few high-res reviews (incl. a video):
- https://www.audiosciencereview.com/...son-linn-records-free-high-res-samples.23366/
- https://www.audiosciencereview.com/...n-high-resolution-music-analysis-video.31620/

 

[RandomEar]

Those peaks in the ultrasonic range do look like filtering artifacts to me, likely from noise shaping.

By increasing the "size" parameter (number of FFT samples), you increase the frequency resolution of your transform. This will increase the fidelity of the plot and extend it to lower frequencies. It will not alter the amplitudes / dB levels of "real signals", but it will change how noise is displayed. This is called FFT gain and is probably what changed your plots. Essentially, you gain information by increasing the number of FFT samples per bin, leading to a better noise rejection and more reliable information about non-noise signals in the bin.

 

[Boye]

Those peaks in the ultrasonic range do look like filtering artifacts to me, likely from noise shaping.

By increasing the "size" parameter (number of FFT samples), you increase the frequency resolution of your transform. This will increase the fidelity of the plot and extend it to lower frequencies. It will not alter the amplitudes / dB levels of "real signals", but it will change how noise is displayed. This is called FFT gain and is probably what changed your plots. Essentially, you gain information by increasing the number of FFT samples per bin, leading to a better noise rejection and more reliable information about non-noise signals in the bin.

Thanks. I guess that means that analysing the files using the default setting - as I did at first - might be somewhat misleading?

 

[OldHvyMec]

"loud frequencies above 20 kHz (ultrasound) can have significant effects on animals, including physiological stress, behavioral changes, and in some cases, physical harm. This is because many animals, unlike humans, can hear in the ultrasonic range.
The specific effects vary depending on the animal's species, age, and individual sensitivity, as well as the intensity, frequency, and duration of the sound."

This includes Dogs, Cats, Rabbits and Rodents.

In other words, it keeps the bugs away and keeps dogs and cats pissed off!

Regards

 

[kyuu]

Garbage and noise is largely what you're paying for with so-called hi-res audio.

 

[krabapple]

Being curious as to why some people claim that hi-res audio files "sound better" than their redbook counterparts I tried to compare them using Audacity (as it is free to use).

Oh dear...are you suggesting a 'hi rez' version might not have higher objective 'audiophile quality' than its plain old 16/44 counterpart?



(I was posting examples of this back to AVSforum in the early 2000s -- in that case it was 'hi rez' masterings that were, to say the least, a bit smashed, dynamically. And cases where the substantial audio content was pretty obviously bounded by ~22kHz -- i.e., the source was "CD rate", the rest is noise. 'Hi rez' for consumer audio has been an effin' scam since, well, forever. It's the mastering that matters, always has, always will.)

 

[RandomEar]

Thanks. I guess that means that analysing the files using the default setting - as I did at first - might be somewhat misleading?

There isn't really a "correct" setting for the FFT samples, same for other stuff like window type. They all depend on the use case.

I'd argue that in this case, a smaller sample size is more helpful or gives you more relevant information than a large one. Because that ultrasonic noise is there and a super large sample size seems to hide it from the plot. And super high spectral resolution isn't that useful when looking at averages of full songs.

 

[somebodyelse]

Oh dear...are you suggesting a 'hi rez' version might not have higher objective 'audiophile quality' than its plain old 16/44 counterpart?



(I was posting examples of this back to AVSforum in the early 2000s -- in that case it was 'hi rez' masterings that were, to say the least, a bit smashed, dynamically. And cases where the substantial audio content was pretty obviously bounded by ~22kHz -- i.e., the source was "CD rate", the rest is noise. 'Hi rez' for consumer audio has been an effin' scam since, well, forever. It's the mastering that matters, always has, always will.)

'Might' being the operative word - you have to take it on a case by case basis as the 'hi rez' might be better, essentially the same, or worse. As you say it's the mastering that matters, and until you check you won't know whether either of them are messed up.

 

[krabapple]

Case by case, it remains true that higher audible quality is never due to 'high rez' consumer formatting per se.

Buying any new/re mastering, regardless of its 'rez' marketing, is a crapshoot.

This is in bold contrast to the claims of many, many hucksters for consumer 'hi rez', going all the way back to Robert Stuart of Meridian.

 

[Boye]

There isn't really a "correct" setting for the FFT samples, same for other stuff like window type. They all depend on the use case.

I'd argue that in this case, a smaller sample size is more helpful or gives you more relevant information than a large one. Because that ultrasonic noise is there and a super large sample size seems to hide it from the plot. And super high spectral resolution isn't that useful when looking at averages of full songs.

Well, this is definitely outside my scope of understanding. As I 'just' want a true or at least relevant representation of what the file contains, I get a little confused that it depends on some settings I really don't understand the meaning of, but thanks for trying

 

[Boye]

Oh dear...are you suggesting a 'hi rez' version might not have higher objective 'audiophile quality' than its plain old 16/44 counterpart?

I sure hope I didn't suggest that. Objectively (not audibly) hi-res might be able to represent ultrasonic sounds like the sounds of bats and even some musical overtones more accurately than 16/44. My question was rather "What is it that I see in the ultrasonic range on the spectrum plots". In the case of bat sounds, it might be actual signals, but as I have seen in demonstrations of "vinyl's extended frequency range" it could just be noise, or it could be other kinds of artefatcs. As it appears in the plots as a separate hump of supersonic information my guess is that it is not just an estended frequency range, but rather unwanted signals caused by aliasing or noise shaping. Who would want that in their system. At best it's a waste of data, but it might even cause problems in the audible range in less than optimal circuitry.

 

[somebodyelse]

Our host has done a series of threads looking at what you might find in high-res recordings, both good and bad:
https://www.audiosciencereview.com/...ic&c[title_only]=1&c[users]=amirm&o=relevance

 

[krabapple]

I sure hope I didn't suggest that.

I hope you did.

(My question was sarcastic, which I hoped would be obvious from the rest of the comment)

That answer to your question about ultrasonic humps in the frequency profile was that if it's there it was part of the 'high rez' creation process, particularly if DSD is involved. Noise in the audible range is 'shaped' into the ultrasonic range. Old school SACD players had filters just before output to get rid of a lot of (but not all) of that hash. Maybe they still do, for all I know.

 

[DVDdoug]

I'm thinking that if you up-sample the low resolution file you should get an apples-to-apples comparison in the FFT/spectrum.

Some of this will be redundant...

When there is an obvious difference, the first thing I suspect is different mastering. Or it could be a different mix or a completely different recording. The high-resolution version might be better, or if it's newer it might be a victim of The Loudness War and sound worse! (Or you might prefer the more dynamically compressed loudness war version.)

Just choose the version that sounds best to you!

The guys at HydrogenAudio have pretty-much demonstrated that under normal conditions you won't hear a difference between a high resolution original and a copy down-sampled to "CD quality" in a proper blind ABX test.

In fact, you normally have to listen very carefully to hear the difference between a high resolution original and a good quality MP3 copy, if you can hear a difference at all. MP3 is lossy and it has a bad reputation (and it can sound terrible at low bitrates) but if you do proper blind listening tests you'll find that it's not as bad as it's reputation.

If you just want to experiment and compare formats without worrying other differences, you can downsample in Audacity. Then if you hear a difference, you really should do a controlled-blind ABX test because it's easy to fool yourself.

DSD does contain ultrasonic noise. It's the nature of the format. You shouldn't hear it and most of it should be filtered-out in an SACD player. If you convert it yourself it may not be properly filtered.

But as I have seen in demonstrations of "vinyl's extended frequency range" it could just be noise

Noise is the "rumor". Maybe distortion too. I'm not sure, but if it's ultrasonic, by definition it's NOT "sound".

You could do a vinyl experiment in Audacity... High pass everything around 20kHz so there's nothing left in the audio range. Then listen to make sure you can't hear anything, and filter again if necessary. Now change the pitch down by half (or more) and listen to what it sounds like. You'll probably want to Amplify too.

The important thing is that digital is ruler-flat across the audio range (at "CD quality" or better). Records and cartridges can extend higher into the ultrasonic but they are NOT as-flat in the important audible range.