The comedy of some Hi res recordings

[pma]

And this is not a 'comedy'. This is a comparison what you get if you buy 96/24 flac (bought from Presto classical) compared to the same title in iTunes.

96/24 flac

iTunes m4a

This + badly treated transients by the lowres lossy format. Depends on music we listen to.

 

[Blumlein 88]

And this is not a 'comedy'. This is a comparison what you get if you buy 96/24 flac (bought from Presto classical) compared to the same title in iTunes.

View attachment 94497

View attachment 94498
96/24 flac

View attachment 94499
iTunes m4a

This + badly treated transients by the lowres lossy format. Depends on music we listen to.

I'd be more bothered by the 300 microphones DG probably used on that recording. Anybody thinking such recording preserves transients with high sample rates should reconsider.

 

[pma]

I'd be more bothered by the 300 microphones DG probably used on that recording. Anybody thinking such recording preserves transients with high sample rates should reconsider.

Spectrum is rich with spectral content up to 40kHz and it is not just noise. As of duality of frequency and time domains, there must be transients captured that are lost with m4a or CD. The diffmaker difference, which is easy to make in this case, shows where is the difference. I am really not a friend of compromises in BW.

 

[Frank Dernie]

Spectrum is rich with spectral content up to 40kHz and it is not just noise. As of duality of frequency and time domains, there must be transients captured that are lost with m4a or CD. The diffmaker difference, which is easy to make in this case, shows where is the difference. I am really not a friend of compromises in BW.

The thing is the content above 20kHz is around 40dB lower than the level at middle frequencies whereas the ears are about 20dB less sensitive at 18kHz (and getting increasingly less sensitive, obviously since we can't hear above 20kHz regardless of how loud it is).
So even if you could hear frequencies above 20kHz this little bit of content would also be too quiet to hear anyway (and is on every frequency content analysis of music I have so far seen).
So having that teensy bit of recording pointless at two levels, frequency and loudness.
When you consider the file size has to be twice as big to record it (and will usually be 24 bit too making it bigger still it really is laughable, so yes very funny, making a file 2.5 times bigger to include a tiny amount of inaudible content that nobody would even be aware was there without a computer to plot it!

 

[Harmonie]

The thing is the content above 20kHz is around 40dB lower than the level at middle frequencies whereas the ears are about 20dB less sensitive at 18kHz (and getting increasingly less sensitive, obviously since we can't hear above 20kHz regardless of how loud it is).
So even if you could hear frequencies above 20kHz this little bit of content would also be too quiet to hear anyway (and is on every frequency content analysis of music I have so far seen).
So having that teensy bit of recording pointless at two levels, frequency and loudness.
When you consider the file size has to be twice as big to record it (and will usually be 24 bit too making it bigger still it really is laughable, so yes very funny, making a file 2.5 times bigger to include a tiny amount of inaudible content that nobody would even be aware was there without a computer to plot it!

I didn't dare, but happy that @Frank Dernie did it for me

I would be curious to see the comparison to an ordinary CD, but couldn't even find the #DG number ..
https://www.deutschegrammophon.com/en/catalogue/products/mussorgsky-pictures-at-an-exhibition-5373

Personally, I'm happy to have all the "bits" and pieces in and prefer the whole "picture" ..
But if we follow the spirit of ASR, who "cares" what's above 20Khz
A small and handy file, a larger file, a spindle ?
Where is the truth ?

 

[Blumlein 88]

Spectrum is rich with spectral content up to 40kHz and it is not just noise. As of duality of frequency and time domains, there must be transients captured that are lost with m4a or CD. The diffmaker difference, which is easy to make in this case, shows where is the difference. I am really not a friend of compromises in BW.

My point is suppose you have even 12 microphones capturing the sound. Instruments are varying distances from the mics and when you combine the 12 feeds any transient will not be a sharp clean transient anymore. It will be a messy out of phase mix of those 12 microphone signals. You'll have signal with high frequency content, but something like a sharp transient of a cymbal being struck will be rather mangled.

I don't know about the particular DG recording referenced. At one time they were infamous for using a forest of microphones. They may not do that now, but I bet it wasn't a purist recording by any means. The idea some clean sharp transients are in the recording is most likely fantasy.

 

[Martin_320]

I concur. I have some Karajan-produced Deutsche Gramophonen "digital recording" CDs from the mid-1980s. And without exception they sound like a mushy mangled mess.

 

[N Fowleri]

First, I want to be clear that I am not trying to provoke heated argument, but appreciate that this is a controversial topic. Second, I am only interested in the results of properly-executed, methodologically-sound, statistically-valid double-blind experiments. Third, I am aware that some purported differences are worthwhile, but they are due to different mastering or artifacts of other differences, as opposed the actual Hi-Res vs. Redbook CD bit-depth and sample frequency. Fourth, I am also aware that just because some people can differentiate the two, it doesn't mean my over-50 ears can.

With that preamble, I saw a large meta-analysis from 2016 that showed that some people can beat random chance differentiating Hi-Res audio, but was it done well? Meta-analyses can be flawed, just like any other research. Also, I know that a novice is not in a good position to judge whether specialized research valid. (Though a novice can sometimes see when a study is clearly not valid.)

As a bonus follow-up, not necessarily related to Hi-Res audio, though I've seen it discussed relative to enhanced up-sampling, is there currently any Artificial Intelligence or other advanced software that can improve audio files. Just as with computational photography, you can't guarantee that you're putting back something that was present in the file, but you can add something in based on knowledge of the world previously acquired.

I look forward to hearing what people have to say.

As full disclosure, I do own a small amount of hi-res audio, including DSD. I sometimes believe I can hear the difference, but I know that doesn't say anything about whether there is--either because of my expectations or because of artifacts in the process. Regarding AI to improve files, I doubt there is anything too magical, or I probably would have heard of it already.

 

[BDWoody]

I think you missed the fact that my link was to the answer to your question about the study...

See post #224

 

[BDWoody]

Regarding AI to improve files, I doubt there is anything too magical, or I probably would have heard of it already.

Sampling Theory really gives you all the magic you need, as it allows the reconstruction of the sampled band limited waveform with no loss of information.

Higher sample rates give you the ability to capture higher frequencies, not recreate anything more accurately or fill in otherwise missing info.

Edit:
If you aren't familiar with sampling theory, have some Monty:

 

[N Fowleri]

I think you missed the fact that my link was to the answer to your question about the study...

See post #224

Yes, I did miss that, sorry. I have now read post #224. Yes, it is nearly a complete answer to my main question. However, it has been almost two years, any new research out?

Sampling Theory really gives you all the magic you need, as it allows the reconstruction of the sampled band limited waveform with no loss of information.

Higher sample rates give you the ability to capture higher frequencies, not recreate anything more accurately or fill in otherwise missing info.

The second question wasn't just about using AI for up-sampling beyond 16/44.1, it was about fixing any type of deficiencies, subjectively or objectively, in the original recording--everything from low-quality microphones to other technical limitations of the the recording or master to fingernails clicking on a piano keyboard (if one doesn't like that sort of thing.)

 

[BDWoody]

Yes, I did miss that, sorry. I have now read post #224. Yes, it is nearly a complete answer to my main question. However, it has been almost two years, any new research out?

I haven't seen any new actual studies. I don't specifically track it, but I'm guessing anything relevant would have made some waves around here. Given the incentive to show it is indeed relevant to music, the lack of support tells its own story to me. I would love to see it, but would bet a lot against it.

The second question wasn't just about using AI for up-sampling beyond 16/44.1, it was about fixing any type of deficiencies, subjectively or objectively, in the original recording--everything from low-quality microphones to other technical limitations of the the recording or master to fingernails clicking on a piano keyboard (if one doesn't like that sort of thing.)

Ah, I'm sorry. You mean something like the software used to take clicks and pops out of records? Other than manually addressing those things you mention individually, I don't know about AI software that would just broadly handle that, but we have plenty of people who work in the field so maybe they will chime in.

 

[DonR]

Once you can establish a pattern for something, you can amplify it or nullify using AI. I would suppose a lot of these sounds follow similar patterns in time and frequency.

 

[RandomEar]

As a bonus follow-up, not necessarily related to Hi-Res audio, though I've seen it discussed relative to enhanced up-sampling, is there currently any Artificial Intelligence or other advanced software that can improve audio files. Just as with computational photography, you can't guarantee that you're putting back something that was present in the file, but you can add something in based on knowledge of the world previously acquired.

The only software I'm currently aware of actually removes unwanted stuff from audio based on "AI": RTX Voice. AMD also announced something similar a couple of days ago. Both aren't really meant for music, though. I think there's also audio editing software which allows you to tune individual instrumental notes in audio files for editing purposes based on "AI". I'm using "AI" here, because you rarely know what's really behind it: Old-school but well tuned deterministic algorithms? Machine learning? Neural networks? Typically, only the developers know the details.