Objectivists - what might be wrong with this label/viewpoint!!

[Don Hills]

I have seen some other research on this beyond the citation of Jkeny. ...

Thank you to both of you.
EDIT: Quite a rabbit hole. Lots of holiday reading, especially Hartmann and Constan. Basically, once you add a room (reverberant field), things get messy fast.

... Now if we did actual soundfield reconstruction we would have something.

I believe JJ has been working on that, based on the hints he's dropped from time to time.
I've heard enough music in 5.1 format to have decided(*) that I'm not going to spend too much effort optimising 2.0 / 2.1 reproduction. Putting my efforts into 5.1 for now (looking at Atmos next) is a more satisfying use of my time/money/energy.

(*) I realised I'm buying much more music in 5.1 format than stereo format these days. While I still have a large legacy stereo collection, I mostly listen to it with headphones.

 

[arnyk]

Compression artifacts in an MP3 encoder can be well above threshold of hearing yet you never see measurements of such Arny.

Perhaps you Amir haven't seen them but I have done such measurements and I have seen them. You just have to measure the right thing which is usually few if any of the exact familiar measures that are used for traditional kinds of audio gear.

Ignores the fact that there are wholly measurement-based techniques for judging MP3 and other perceptual decoders. Here is just a few examples of many:

http://archive.arstechnica.com/wankerdesk/1q00/mp3/mp3-2.html

http://en.wikipedia.org/wiki/PEAQ AKA ITU-R Recommendation BS.1387

On balance DBTs are a more generally accepted methodology fro judging coder quality, but when people say that nobody ever measures or has measured MP3 encoders they are speaking out of ignorance and are begging for correction.


Dynamic distortion that comes and goes based on what is played or what the equipment is doing at that moment is a glaring problem with our measurements today.

 

[jkeny]

I have seen some other research on this beyond the citation of Jkeny. They have a bit of a different theory on what is happening. Related to pulse trains of higher frequency harmonically related tones. Paraphrasing from memory the related trains of pulses with a varying amplitude were treated somewhat as if the frequency of the amplitude changes were what is being timed. Like the Jkeny citation, they too found the threshold 10 or more times higher. So ITD can be used for localization less precisely on some types of signals.

Thanks, esl, I would be interested in reading these if you can find the articles.

 

[jkeny]

...
Dynamic distortion that comes and goes based on what is played or what the equipment is doing at that moment is a glaring problem with our measurements today.

So, by this you mean that dynamic music-like signals with realistic crest factors that stress equipment in ways which steady state periodic signals don't, is a useful test but any dynamic distortions arising from such tests "is a glaring problem with our measurements today" & therefore such tests are not done.

 

[arnyk]

So, by this you mean that dynamic music-like signals with realistic crest factors that stress equipment in ways which steady state periodic signals don't, is a useful test but any dynamic distortions arising from such tests "is a glaring problem with our measurements today" & therefore such tests are not done.

Obviously not that, because it is generally not true.

Dynamic music-like signals with realistic crest factors don't generally stress equipment in ways which steady state periodic signals don't. In fact the opposite is more generally true. Steady state signals generally produce more stress since they require handing more energy on the average, to reproduce properly.

For example music has a crest factor that runs from about 6 dB to over 20 dB. This means that its average power is far less than that of a steady state sine wave with the peak value. Average power is what cooks output transistors, drains power supplies, and fries speaker drivers.

So, we are back to the objectivist truism that if a piece of gear can handle a steady state test signal with the required peak value without clipping, handling real world music is usually a slam dunk.

 

[jkeny]

Obviously not that, because it is generally not true.

Dynamic music-like signals with realistic crest factors don't generally stress equipment in ways which steady state periodic signals don't. In fact the opposite is more generally true. Steady state signals generally produce more stress since they require handing more energy on the average, to reproduce properly.

Have you got any measurements that backs up this claim?

For example music has a crest factor that runs from about 6 dB to over 20 dB. This means that its average power is far less than that of a steady state sine wave with the peak value. Average power is what cooks output transistors, drains power supplies, and fries speaker drivers.

What bit about "DYNAMIC non-periodic signals" do you not understand?

So, we are back to the objectivist truism that if a piece of gear can handle a steady state test signal with the required peak value without clipping, handling real world music is usually a slam dunk.

A claim that any objectivist should be able to back up with measurements not just truisms.

 

[amirm]

Perhaps you Amir haven't seen them but I have done such measurements and I have seen them. You just have to measure the right thing which is usually few if any of the exact familiar measures that are used for traditional kinds of audio gear.

Let's see you post them Arny.

Ignores the fact that there are wholly measurement-based techniques for judging MP3 and other perceptual decoders. Here is just a few examples of many:

http://archive.arstechnica.com/wankerdesk/1q00/mp3/mp3-2.html

Don't care about any online examples from bloggers. Yes, you can measure the frequency response of the encoder. No, that doesn't tell anything about non-linear distortions that are dynamic as I pointed out. No static snapshot tell you that.

http://en.wikipedia.org/wiki/PEAQ AKA ITU-R Recommendation BS.1387

That is not an audio measurement as you post. It is a perceptual analysis method. While many of these have been developed, they play little if any role in development of lossy audio compression. Instead, their role is for automated testing to catch software bugs and such that generate really incorrect results.

Now, if you had one of these for measuring audio equipment we would be one step ahead. Instead, you are defending psychoacoustically blind measurements of audio that date decades back and were developed for their simplicity and when such distortions were high.

On balance DBTs are a more generally accepted methodology fro judging coder quality, but when people say that nobody ever measures or has measured MP3 encoders they are speaking out of ignorance and are begging for correction.

If DBTs are the more generally accepted methodology, then by your own statement measurements play little to no role when non-linear distortions are dynamic. Exactly the point I made that you objected to. So no, there was no correction. Just a few empty google search results posted without due knowledge of their applicability.

 

[amirm]

For example music has a crest factor that runs from about 6 dB to over 20 dB. This means that its average power is far less than that of a steady state sine wave with the peak value. Average power is what cooks output transistors, drains power supplies, and fries speaker drivers.

You better be relying on such measurements with real, complex loads simulating speakers, not resistive loads. With non-resistive loads, you can easily cook the amplifier when the same steady state signal using resistive dummy loads would not have.

Difference between real amplifiers, and paper ones you are assuming.

So, we are back to the objectivist truism that if a piece of gear can handle a steady state test signal with the required peak value without clipping, handling real world music is usually a slam dunk.

Per above, nothing is remotely so if the steady state measurement does not match *your* speaker load. Protection circuit in the amplifier will kick in with complex loads and create dynamic, non-linear distortions that do not show up when you feed pretty sine waves to a resistor.

 

[Atmasphere]

Have you considered that the distortion may be emphasising the characteristics of the sources that "define the instrumental timbres" and the "low level detail"? SE tube amplifiers as a class have higher distortion but are often preferred because they "reveal more of the music."

Text in bold: this is entirely subjective, and IMO, false. SETs are known for good low level detail because as power is decreased, distortion drops linearly to unmeasurable. When distortion is removed, detail emerges. As with any amp, an SET spends most of its time at low power due to the nature of music. Generally speaking, the SET is set up incorrectly if more than about 20% of its full power is demanded by the user on peaks.

But I find SETs lacking in detail relative to the amps I use... -which are push pull, and unlike most P-P amps, also have distortion that decreases linearly to unmeasurable. But because they are P-P they have a lot less distortion.

The study of Psychoacoustics is a subset of Engineering so there can never be a proper Either/Or relationship between them.

This is simply untrue. We are talking about humans here, and as such perfectly good engineers often are uninformed about other branches of their practice. I've seen it many times- and we see it in the test and measurement regime, which is based mostly on psychoacoustic knowledge from the 1960s, and even then ignored information known at that time.

Dynamic music-like signals with realistic crest factors don't generally stress equipment in ways which steady state periodic signals don't. In fact the opposite is more generally true.

You didn't really mean to say this as written, correct? What were you actually stating?

 

[Orb]

I have seen some other research on this beyond the citation of Jkeny. They have a bit of a different theory on what is happening. Related to pulse trains of higher frequency harmonically related tones. Paraphrasing from memory the related trains of pulses with a varying amplitude were treated somewhat as if the frequency of the amplitude changes were what is being timed. Like the Jkeny citation, they too found the threshold 10 or more times higher. So ITD can be used for localization less precisely on some types of signals.

I rather doubt this is common. I have tried various simple recordings in mono in various ways over headphones. Simplest is delay one channel versus another for solo violin or viola. Could never get anything, but a mono image in the middle. Do the same, but with a bit of frequency shaping and get some effect slightly to one side. It was very vague, and doing away with the delay and using only frequency shaping on one channel managed about half the effect. Put a level difference on one channel and it was a clear effect over to one side.

Whether this is important or not really isn't a big deal in this context. Regular old redbook standards will encode any of this happening way more than well enough. For that matter so did analog tape. Stereo is an effect, an illusion built upon certain characteristics of human hearing. With simple techniques it can be surprisingly accurate with imaging. Now if we did actual soundfield reconstruction we would have something.

Just to add esldude,
not sure how reliable delaying one channel over headphones as more recent studies has shown how the brain can actually swap left and right side; I think the study utilised loudness and sorry for the really vague post on this but I would need to find the paper.
Basically it relied upon one channel quieter and one louder (using headphones), really do not want to say more than that without re-reading the paper.
Cheers
Orb

 

[jkeny]

For example music has a crest factor that runs from about 6 dB to over 20 dB. This means that its average power is far less than that of a steady state sine wave with the peak value. Average power is what cooks output transistors, drains power supplies, and fries speaker drivers.

You trip yourself up at every post, Arny & demonstrate yet again your lack of technical know-how & logic - the fact that the crest factor for music is far higher than for sinusoidal-like signals is the whole point of the test - to test the electronics ability to faithfully handle such disparity in dynamic power delivery. Your posts show not the slightest notion of what's being talked about!

 

[esldude]

Thanks, esl, I would be interested in reading these if you can find the articles.

Maybe I can locate them. I was researching all the things that can effect our hearing in regard to spatial location. Wondering if any techniques might work better than the standard ones for miking live recordings. The basic phase below 800 hz and intensity above 2 khz still seems the most of it. In looking into it I ran across a couple papers similar to the one you cited. Showing in some conditions groups of higher tones can be somewhat located by phase. I was doing all that 6 months ago, and don't think I kept copies of the papers.

 

[esldude]

Thank you to both of you.
EDIT: Quite a rabbit hole. Lots of holiday reading, especially Hartmann and Constan. Basically, once you add a room (reverberant field), things get messy fast.



I believe JJ has been working on that, based on the hints he's dropped from time to time.
I've heard enough music in 5.1 format to have decided(*) that I'm not going to spend too much effort optimising 2.0 / 2.1 reproduction. Putting my efforts into 5.1 for now (looking at Atmos next) is a more satisfying use of my time/money/energy.

(*) I realised I'm buying much more music in 5.1 format than stereo format these days. While I still have a large legacy stereo collection, I mostly listen to it with headphones.

Unfortunately in the case of JJ, I believe he did work on that while at Bell Labs, and Bell Labs owns the intellectual property. My impression was he had developed an effective practical system for it. He mentioned in a video that for whatever reason ATT/Bell wasn't interested in pursuing it commercially though it would seem to have a potential audience. Maybe Amir knows something of it or JJ might say himself though I am not sure JJ takes part here anymore.

 

[arnyk]

You trip yourself up at every post, Arny & demonstrate yet again your lack of technical know-how & logic - the fact that the crest factor for music is far higher than for sinusoidal-like signals is the whole point of the test - to test the electronics ability to faithfully handle such disparity in dynamic power delivery. Your posts show not the slightest notion of what's being talked about!

My dear man it is you that have taken the face plant, as my comments about high crest factor signals having relatively low amounts of average energy and thus being less stressful for audio gear to amplify cleanly is well known to second year EE students. This is so basic that Amir may even know it. ;-)

Here is a little light reading on the topic for you: http://www.eetimes.com/document.asp?doc_id=1275672

 

[jkeny]

My dear man it is you that have taken the face plant, as my comments about high crest factor signals having relatively low amounts of average energy and thus being less stressful for audio gear to amplify cleanly is well known to second year EE students. This is so basic that Amir may even know it. ;-)

Here is a little light reading on the topic for you: http://www.eetimes.com/document.asp?doc_id=1275672

You still haven't a clue, I see!

 

[JackD201]

We have been getting multiple reports from both sides. Time to decide people. Do you want this thread closed or will you be civil?

 

[JackD201]

They are the people I am addressing Tom. They know who they are.

 

[Don Hills]

Unfortunately in the case of JJ, I believe he did work on that while at Bell Labs, and Bell Labs owns the intellectual property. My impression was he had developed an effective practical system for it. He mentioned in a video that for whatever reason ATT/Bell wasn't interested in pursuing it commercially though it would seem to have a potential audience. Maybe Amir knows something of it or JJ might say himself though I am not sure JJ takes part here anymore.

I have access to a lot of the Bell Labs papers. I'll take a look over the break.

Regarding the MacPherson/Middlebrooks paper, they left the money shot till last:

These simulation results suggest that, in agreement with the hypothesis of Middlebrooks and Green ~1990!, high-frequency Gaussian noise signal envelopes cannot convey robust ITD information.

 

[arnyk]

Let's see you post them Arny.

One of the classic ways to show how perceptual coders trash audio signals is to measure their performance with multitones. Multitones are a closer simulation of music than small numbers of pure tones. They are a test tone that people have been trying to popularize in modern audio measurements for decades:

http://www.ap.com/kb/show/60

Here is a 4416 file of a multitone:



Here is the same audio file after being turned into a 128 K MP3:



There have been a number of dramatic technical degradations of the file when encoded via MP3, the most visible of are:

(1) The noise floor between the tones has increased tremendously - something like 40 dB. Generally this suggests nonlinear distortion but with MP3 encoders it is usually something more complex - it is usually band limited random noise.

(2) A brick wall filter was inserted above about 15.5 KHz.

I posted similar measurements on my www.pcavtech.com web site starting in the middle 1990s.

Clark and others posted similar results for tests at Audio Magazine related to the Philips DCC system when it first came out - when was that? Wkipedia says 1992... 20+ years ago! I believe he also posted the positive results of ABX testing. To the best of my knowledge there was no revolt among objectivists as some might predict - minimal standards for reliable results were followed.

Do Atkinson or Amir post the results of multitone tests done with their fancy AP gear? According to the AP document above they are loaded for bear. If they are so esoteric why is this sad poor stupid amateur posting multitone results with such ease?

I'm just an amateur, guys! ;-)

Old, old, old news.

 

[arnyk]

(snip comments about SETs)

The study of Psychoacoustics is a subset of Engineering so there can never be a proper Either/Or relationship between them.

This is simply untrue. We are talking about humans here.

No we are not. We are talking about a general field of study and a profession. That field is Engineering which is hugely broad. The failings of specific individuals in this field is another matter.

and as such perfectly good engineers often are uninformed about other branches of their practice. I've seen it many times- and we see it in the test and measurement regime, which is based mostly on psychoacoustic knowledge from the 1960s, and even then ignored information known at that time.

Now that is true, and I've seen the same many times and I have lamented about it on various forums.

Dynamic music-like signals with realistic crest factors don't generally stress equipment in ways which steady state periodic signals don't. In fact the opposite is more generally true.

You didn't really mean to say this as written, correct? What were you actually stating?

I mean exactly what I said. It falls out of the very definition of crest factor!

http://en.wikipedia.org/wiki/Crest_factor

"Crest factor is a measure of a waveform, such as alternating current or sound, showing the ratio of peak values to the average value. In other words, crest factor indicates how extreme the peaks are in a waveform. Crest factor 1 indicates no peaks, such as direct current. Higher crest factors indicate peaks, for example sound waves tend to have high crest factors."

Objectivists are commonly faulted for saying things like "If a reasonably good amplifier can handle a signal without clipping it will be free of audible distortion." However, looking at the formal definition of Crest factor we see that the statement presumes that the peaks are handled without distortion. A ton of placebophile hand wringing about the differences between sine waves and music goes right out the window!

I am trying to communicate the truth that a power amp that has to handle a low crest factor signal will generally handle a high crest factor signal with the same peak amplitude as long as it does not clip the peaks. Along the way I observe that the high crest factor signal by definition contains less average energy which makes fewer demands on output transistors, power supplies and heat sinks. This falls out of the very definition of crest factor!

This is readily observable when testing doing sine waves are compared to tests using actual musical waveforms. The music tests draw less current from the power line, the heat up the amp far less, and they are generally easier to complete no matter how long they run.