Skip to main content
Topic: Hi Rez vs Redbook in Classical music (Read 15696 times) previous topic - next topic
0 Members and 1 Guest are viewing this topic.

Hi Rez vs Redbook in Classical music

Reply #100
How about a form of that command that can actually encode an existing file?  Pretty please? ;-)

Code: [Select]
# Widest allowed transition without aliasing; with default dither:
sox in.wav -b16 out1.wav rate -m -b74 44100
# Widest allowed transition with aliasing; with 'shibata' dither:
sox in.wav -b16 out2.wav rate -m -b85 -a 44100 dither -s

In both cases. the passband can be extended by increasing the -b number towards (but not reaching) 100 %

Hi Rez vs Redbook in Classical music

Reply #101
Nothing seems to be able to match a CEP filter with Quality=100.
Link to graphics in uploads

Indeed low ringing and like mentioned before a lowpass kicking in at 14kHz but the aliasing alone is nothing i'd like.
Above is Audition 3 Q=100, below SoX -b92 -a

Is troll-adiposity coming from feederism?
With 24bit music you can listen to silence much louder!

Hi Rez vs Redbook in Classical music

Reply #102
Nothing seems to be able to match a CEP filter with Quality=100.
Link to graphics in uploads

Indeed low ringing and like mentioned before a lowpass kicking in at 14kHz but the aliasing alone is nothing i'd like.
Above is Audition 3 Q=100, below SoX -b92 -a



OK, I won't bore you with pictures but CEP @ Q=150 puts aliasing down about 70 dB @ 22050 Hz, has less than half the ringing of Sox,

It is only 0.1 dB down @ 17.6 KHz and about 17 dB down at 20 KHz. 

If someone wants to try to ABX it, I will provide processed and unprocessed 24/192 or 24/96 files of musical selections that are available to me under reasonable conditions.

Hi Rez vs Redbook in Classical music

Reply #103
In CEP you need the pre/post filter enabled. You get some aliasing if you disable it.

If these kind of trade offs matter, it's interesting to consider how much better even 48kHz is than 44.1kHz. A lot of people have claimed to hear that difference, but ABX results are rather thin on the ground.

Cheers,
David.

Hi Rez vs Redbook in Classical music

Reply #104
OK, I won't bore you with pictures but CEP @ Q=150 puts aliasing down about 70 dB @ 22050 Hz, has less than half the ringing of Sox,

It is only 0.1 dB down @ 17.6 KHz and about 17 dB down at 20 KHz. 

If someone wants to try to ABX it, I will provide processed and unprocessed 24/192 or 24/96 files of musical selections that are available to me under reasonable conditions.

But you realize it has nothing to do with SoX? If you will keep as much frequency content and use exactly as low ringing as any SoX setting the ringing will be the same with CEP.
And no, starting lowpassing at 17kHz with -17dB at 20kHz with many aliasing is not the new philosophers stone.
Is troll-adiposity coming from feederism?
With 24bit music you can listen to silence much louder!

Hi Rez vs Redbook in Classical music

Reply #105
OK, I won't bore you with pictures but CEP @ Q=150 puts aliasing down about 70 dB @ 22050 Hz, has less than half the ringing of Sox,

It is only 0.1 dB down @ 17.6 KHz and about 17 dB down at 20 KHz. 

If someone wants to try to ABX it, I will provide processed and unprocessed 24/192 or 24/96 files of musical selections that are available to me under reasonable conditions.

But you realize it has nothing to do with SoX? If you will keep as much frequency content and use exactly as low ringing as any SoX setting the ringing will be the same with CEP.


I wish that were true or were even a reasonable expectation.

Sox does not seem to allow every anti-aliasing filter possible within its core technology to be defined, but neither does CEP.  Sox's  UI no doubt does not allow fully exploiting the underlying subroutine library and that underlying subroutine library probably does not allow defining every configuration that is possible with the core technology. Such are the practical limits of programming.

Simple challenge - prove me wrong and produce an anti aliasing filter with equal low ringing, intrusion on the audible band, and aliasing with Sox. That would prove your point eloquently.

Quote
And no, starting lowpassing at 17kHz with -17dB at 20kHz with many aliasing is not the new philosophers stone.


Of course not, and as Stefan Heinzmann alludes to in his AES conference post that I have linked and quoted (Heinzmann's latest AES forum post). There are existing hardware implementations that do just that. I seem to recall that the DACs in the Behringer UCA 202 are designed that way, for example.

More relevant info here: http://www.hydrogenaud.io/forums/index.php...st&p=894226

Hi Rez vs Redbook in Classical music

Reply #106
I do enjoy the zeal of the new convert


Hi Rez vs Redbook in Classical music

Reply #108
The AES paper triggerd the "fear of ringing" in you, i guess thats what David means.
Is troll-adiposity coming from feederism?
With 24bit music you can listen to silence much louder!

Hi Rez vs Redbook in Classical music

Reply #109
The AES paper triggered the "fear of ringing" in you, i guess that's what David means.


Actually its "Fear of Snake Oil" and it is very old in me.

There was a bit of opportunism. It gave me an excuse to cut ties with some SR-related  DBTs that had been thoroughly abused by you-know-who.

The paper seemed like a great opportunity to skewer high ranking golden ears masquerading as scientists, and that is always fun.

Other than that there do seem to be a lot of loose ends related to past DBTs in this area.

Hi Rez vs Redbook in Classical music

Reply #110
I do enjoy the zeal of the new convert
It's April after all
I have the impression that looking at the frequency domain won't give meaningful answers about audibility. Hi-res and redbook can be identical below 20 kHz. But looking at the time domain in a DAW has a risk too, since it doesn't apply an auditory filter. Remember amirm saying about (pre-)ringing: "We hear each one of those samples as they come." I remain skeptical about that, but if there's convincing evidence I'll gladly join the converts.
Am I right that most (if not all) of the hi-res tests have been done binaurally (2 ears)? It might be interesting to find out if monotic (1 ear) tests give comparable results. I know from experience that interaural differences can be very revealing, not only for the source but also for additional distortion.

Hi Rez vs Redbook in Classical music

Reply #111
Remember amirm saying about (pre-)ringing: "We hear each one of those samples as they come." I remain skeptical about that, but if there's convincing evidence I'll gladly join the converts.

amirm's uttering is of course obvious bullshit. We don't hear the samples at all, we hear a reconstruction, which is a time-continuous waveform. If the reconstruction was faithful enough, we can't distinguish the result from the original.

If the reconstruction contains pre- or post-echos which weren't there before, the result may be audibly different. Because of masking, the effect will be much more critical with the pre-echo. I think this much should be uncontroversial.

The frequency response measurements may not show such artefacts, because the measurement typically is done in a steady-state condition, where the echos aren't there. Such effects may therefore have to be measured in a different way.

This is actually quite an old hat. The last reference in the Meridian paper is a paper by Lagadec and Stockham from 1984, where they found audible effects related to pre-echos in a somewhat extreme and contrived filter bank, which wouldn't have been apparent from the frequency response. They found it correlated with the passband ripple of the filter, which consequently has become one of the characterizing data items found in the data sheets of audio converters.

None of this means that we need higher sampling rates. It just means that there is more to a filter than its frequency response. Maybe there are subtle things we have yet to learn about them, but I'm not an expert there. It is intuitively clear, however, that a steeper filter will (in the case of FIR) have more taps, and hence a potentially longer pre-echo (depending on its design). At some point it might become audible. That's a question that can't be answered easily from impulse response diagrams.

Anyway, if you know the problem, and know how to determine its magnitude, you can design the filter to avoid it. As is often the case in engineering, this may require a compromise between several conflicting requirements. Perhaps higher sampling rates make that compromise easier to reach (or to avoid), but at a disproportionate cost. I'm still not convinced this is necessary. In fact, I am less convinced today than 15 years ago. The completely overblown and nonsensical claims of hi-res audibility have left their mark. I believe that while 44.1 kHz may be a bit tight, 48 kHz ought to be sufficient. That's more a gut feeling I've developed over a couple of decades, rather than anything based on solid, irrefutable evidence. But in my mind, studies such as the Meridian paper actually confirm my position rather than challenge it, because it is apparent that audibility was at the limit of detection, which is pretty much what was to be expected, give or take a bit.

Hi Rez vs Redbook in Classical music

Reply #112
Remember amirm saying about (pre-)ringing: "We hear each one of those samples as they come."


As I see it this was based on a misunderstanding of something JJ once wrote in the context of the cochlear filters.


As an aside: ringing becomes audible when the transition band gets narrower than the ear's critical band width at the transition frequency, give or take. This can be demonstrated easily around 3kHz.
This also suggests that for CD-rate anti-aliasing a transition band of 4kHz suffices, i.e. starting the rolloff at 18kHz.
If this is done at the production side then the replay side can happily keep using the somewhat compromised half-band filters that are deployed in just about every mass-market DAC chip in use today.


Hi Rez vs Redbook in Classical music

Reply #113
As an aside: ringing becomes audible when the transition band gets narrower than the ear's critical band width at the transition frequency, give or take. This can be demonstrated easily around 3kHz.

Do you have a reference for this?


Hi Rez vs Redbook in Classical music

Reply #115
Remember amirm saying about (pre-)ringing: "We hear each one of those samples as they come." I remain skeptical about that, but if there's convincing evidence I'll gladly join the converts.

amirm's uttering is of course obvious bullshit.


Agreed, and what's new?

The old joke about how to tell a salesman or a politician is lying comes to mind. ;-)

Quote
We don't hear the samples at all, we hear a reconstruction, which is a time-continuous waveform.


Above a certain fairly low frequency < 1 KHz we don't hear the waveform at all. In very general terms we hear its spectral analysis.

Quote
The frequency response measurements may not show such artefacts, because the measurement typically is done in a steady-state condition, where the echos aren't there. Such effects may therefore have to be measured in a different way.


Not true of most frequency analysis done these days. We generally measure frequency response using FFTs, and a FFT has a defined area of interest in the time domain based on the length and the windowing of the sampled data.  The data may be averaged, but that is up to the person performing the analysis.


Quote
This is actually quite an old hat. The last reference in the Meridian paper is a paper by Lagadec and Stockham from 1984, where they found audible effects related to pre-echos in a somewhat extreme and contrived filter bank, which wouldn't have been apparent from the frequency response. They found it correlated with the passband ripple of the filter, which consequently has become one of the characterizing data items found in the data sheets of audio converters.


Interesting. I was wondering where the obsession with passband ripple comes from. No  matter, even cheap converters have passband ripple that is infinitesimal.

Quote
None of this means that we need higher sampling rates. It just means that there is more to a filter than its frequency response. Maybe there are subtle things we have yet to learn about them, but I'm not an expert there. It is intuitively clear, however, that a steeper filter will (in the case of FIR) have more taps, and hence a potentially longer pre-echo (depending on its design). At some point it might become audible. That's a question that can't be answered easily from impulse response diagrams.


Agreed.

Quote
Anyway, if you know the problem, and know how to determine its magnitude, you can design the filter to avoid it. As is often the case in engineering, this may require a compromise between several conflicting requirements. Perhaps higher sampling rates make that compromise easier to reach (or to avoid), but at a disproportionate cost. I'm still not convinced this is necessary. In fact, I am less convinced today than 15 years ago. The completely overblown and nonsensical claims of hi-res audibility have left their mark. I believe that while 44.1 kHz may be a bit tight, 48 kHz ought to be sufficient. That's more a gut feeling I've developed over a couple of decades, rather than anything based on solid, irrefutable evidence. But in my mind, studies such as the Meridian paper actually confirm my position rather than challenge it, because it is apparent that audibility was at the limit of detection, which is pretty much what was to be expected, give or take a bit.


What the golden ears don't seem to want to answer for is their decade-plus of falling prostrate before alleged high-rez recordings that for certain had low-rez provenance.  Once resolution is gone you can't put it back, not even with $10,000 DACs and $100,000 speakers.

Hi Rez vs Redbook in Classical music

Reply #116
If the reconstruction contains pre- or post-echos which weren't there before, the result may be audibly different. Because of masking, the effect will be much more critical with the pre-echo. I think this much should be uncontroversial.

The frequency response measurements may not show such artefacts, because the measurement typically is done in a steady-state condition, where the echos aren't there. Such effects may therefore have to be measured in a different way.

This is actually quite an old hat. The last reference in the Meridian paper is a paper by Lagadec and Stockham from 1984, where they found audible effects related to pre-echos in a somewhat extreme and contrived filter bank, which wouldn't have been apparent from the frequency response. They found it correlated with the passband ripple of the filter, which consequently has become one of the characterizing data items found in the data sheets of audio converters.

None of this means that we need higher sampling rates. It just means that there is more to a filter than its frequency response. Maybe there are subtle things we have yet to learn about them, but I'm not an expert there. It is intuitively clear, however, that a steeper filter will (in the case of FIR) have more taps, and hence a potentially longer pre-echo (depending on its design). At some point it might become audible. That's a question that can't be answered easily from impulse response diagrams.
All this applies to filters in the audible range. That's old hat. (btw, you could look at a given impulse response and determine whether the pre-ringing from it was audible. There are known temporal masking curves, just like there are known spectral masking curves.)

What would be significant would be proving beyond doubt that it applies to filters operating only at ultrasonic filters, and determining why.


Quote
Anyway, if you know the problem, and know how to determine its magnitude, you can design the filter to avoid it. As is often the case in engineering, this may require a compromise between several conflicting requirements. Perhaps higher sampling rates make that compromise easier to reach (or to avoid), but at a disproportionate cost. I'm still not convinced this is necessary. In fact, I am less convinced today than 15 years ago. The completely overblown and nonsensical claims of hi-res audibility have left their mark. I believe that while 44.1 kHz may be a bit tight, 48 kHz ought to be sufficient. That's more a gut feeling I've developed over a couple of decades, rather than anything based on solid, irrefutable evidence. But in my mind, studies such as the Meridian paper actually confirm my position rather than challenge it, because it is apparent that audibility was at the limit of detection, which is pretty much what was to be expected, give or take a bit.
I agree with that last part.

However, I don't see higher sample rates in themselves being a disproportionate cost. Using higher sample rates at home doesn't cost me anything. If it makes a difference - and I mean any positive difference - I'll use them.

It's the whole "buy into this - you'll have to pay much more for the equipment and recordings, but you'll hear a night-and-day improvement" that's wrong.

It only makes any sense at all if you can hear a difference though.


Should we now follow the new conventional wisdom and use today's best compromise gentle filtering? Here's some food for thought: If you'd listened to conventional HA wisdom 5 years ago, you'd have created and replayed recordings at 44.1kHz 16-bit with the steepest possible anti-alias and anti-imaging filters. This practice might, it seams, introduce a just-ABXable change in the sound. If you'd ignored conventional wisdom and used 24/96, you would have avoided this just-ABXable change in the sound. Who was right back then?  Who is right today?

I realise this argument leads to 28MHz 64-bit audio sampling, and madness. I'm not seriously making it, because I still record most analogue sources to 44.1kHz/16-bit (if the levels are well controlled) or 48kHz/24-bit (if the levels are badly controlled and it's for a video sound track). Even so, it should make you wonder.

Cheers,
David.

Hi Rez vs Redbook in Classical music

Reply #117
However, I don't see higher sample rates in themselves being a disproportionate cost. Using higher sample rates at home doesn't cost me anything. If it makes a difference - and I mean any positive difference - I'll use them.

I certainly don't object to that. I wasn't referring to the situation in a recording, production or mastering studio, but in audio distribution. There, size still matters, and cost can't be ignored. Yet I agree with you that the main problem is the marketing promises which are wrong.

Quote
Should we now follow the new conventional wisdom and use today's best compromise gentle filtering? Here's some food for thought: If you'd listened to conventional HA wisdom 5 years ago, you'd have created and replayed recordings at 44.1kHz 16-bit with the steepest possible anti-alias and anti-imaging filters. This practice might, it seams, introduce a just-ABXable change in the sound. If you'd ignored conventional wisdom and used 24/96, you would have avoided this just-ABXable change in the sound. Who was right back then?  Who is right today?

I don't think that the right answer to that question is known yet. If someone knows it, please step forward. However, even if there are very small differences in sound because of this, I am quite sure that they do not matter for the enjoyment or quality of a recording that is on sale. If it were, it would be open to the producer to select the best filtering for his product, because it is something that happens on the production side, namely at the point where the sampling rate and wordlength are reduced for the final product. So it merely becomes a quality feature of the individual product, and no new format would have to be forced into the market.

Quote
I realise this argument leads to 28MHz 64-bit audio sampling, and madness. I'm not seriously making it, because I still record most analogue sources to 44.1kHz/16-bit (if the levels are well controlled) or 48kHz/24-bit (if the levels are badly controlled and it's for a video sound track). Even so, it should make you wonder.

If the madness were confined to your own studio, I wouldn't even object. If you believe that 28 MHz/64-bit helps you make better recordings, fine. I would doubt it, but it is you who has to deal with the ramifications, and as long as the outcome is a good product in the "normal" format, that's ok with me. I take it for granted anyway, that many shops are working with 96/24. There's some sense in that, quite remote from audibility discussions. At the point when it hits distribution, the situation is different, however, and the benefit disappears.

 
SimplePortal 1.0.0 RC1 © 2008-2019