IM Distortion, Speakers and the Death of Science


One topic that often comes up is perception vs. measurements.

"If you can't measure it with common, existing measurements it isn't real."

This idea is and always will be flawed. Mind you, maybe what you perceive is not worth $1, but this is not how science works. I'm reminded of how many doctors and scientists fought against modernizing polio interventions, and how only recently did the treatment for stomach ulcers change radically due to the curiosity of a pair of forensic scientists.

Perception precedes measurement.  In between perception and measurement is (always) transference to visual data.  Lets take an example.

You are working on phone technology shortly after Bell invents the telephone. You hear one type of transducer sounds better than another.  Why is that?  Well, you have to figure out some way to see it (literally), via a scope, a charting pen, something that tells you in an objective way why they are different, that allows you to set a standard or goal and move towards it.

This person probably did not set out to measure all possible things. Maybe the first thing they decide to measure is distortion, or perhaps frequency response. After visualizing the raw data the scientist then has to decide what the units are, and how to express differences. Lets say it is distortion. In theory, there could have been a lot of different ways to measure distortion.  Such as Vrms - Vrms (expected) /Hz. Depending on the engineer's need at the time, that might have been a perfectly valid way to measure the output.

But here's the issue. This may work for this engineer solving this time, and we may even add it to the cannon of common measurements, but we are by no means done.

So, when exactly are we done?? At 1? 2? 5?  30?  The answer is we are not.  There are several common measurements for speakers for instance which I believe should be done more by reviewers:

- Compression
- Intermodulation ( IM ) Distortion
- Distortion

and yet, we do not. IM distortion is kind of interesting because I had heard about it before from M&K's literature, but it reappeared for me in the blog of Roger Russel ( http://www.roger-russell.com ) formerly from McIntosh. I can't find the blog post, but apparently they used IM distortion measurements to compare the audibility of woofer changes quite successfully.

Here's a great example of a new measurement being used and attributed to a sonic characteristic. Imagine the before and after.  Before using IM, maybe only distortion would have been used. They were of course measuring impedance and frequency response, and simple harmonic distortion, but Roger and his partner could hear something different not expressed in these measurements, so, they invent the use of it here. That invention is, in my mind, actual audio science.

The opposite of science would have been to say "frequency, impedance, and distortion" are the 3 characteristics which are audible, forever. Nelson pass working with the distortion profile, comparing the audible results and saying "this is an important feature" is also science. He's throwing out the normal distortion ratings and creating a whole new set of target behavior based on his experiments.  Given the market acceptance of his very expensive products I'd say he's been damn good at this.

What is my point to all of this?  Measurements in the consumer literature have become complacent. We've become far too willing to accept the limits of measurements from the 1980's and fail to develop new standard ways of testing. As a result of this we have devolved into camps who say that 1980's measures are all we need, those who eschew measurements and very little being done to show us new ways of looking at complex behaviors. Some areas where I believe measurements should be improved:

  • The effects of vibration on ss equipment
  • Capacitor technology
  • Interaction of linear amps with cables and speaker impedance.

We have become far too happy with this stale condition, and, for the consumers, science is dead.
erik_squires
That andy2 and mahgister and you kevin repeatedly describe it purely as algorithmic, i.e. the same data will always result in the exactly same answer to n-decimal places, clearly communicates that your knowledge of AI is rudimentary at best and hence you type long posts on an audio forum site repeatedly that are fundamentally wrong as opposed to seeking out the knowledge that you lack.
I dont know for Andy2, but you cannot presume at all of what I understand and what not sorry..... I post to you a conference by Naftaly Tishby, that is not for the lay man and describe neural networks mathematical workings then i never describe the algorithm of neural networks like you said I did in misrepresenting me.... Have you read my posts?

Our biggest advantage is we achieved enough intelligence to enable formal communication
By the way this affirmation is totally ingenuous to say the least....

Read Ernst Cassirer a true scientist and philosopher to understand why it is naïve and totally superficial affirmation....After that you will understand the difference in concept "creation" in neural networks and true concepts creation....

You "believe" you are right, and hence, as opposed to learning, you keep repeating the same mistakes.
You are polite and thanks for that... But you are also arrogantly patronizing us about knowledge without knowing to who you talks... :) I give you some hint when I gives you names like Naftaly Tishby or Grothendieck…. Do you even know the names?
Post removed 
heaudio123"The crux of the discussion kevn (and andy2 and mahgister to some degree) is that you have no concept of how AI works, or can work, or how it is or can be implemented."

Actually what you’re remarks display and reveal and I will not quote them all here is that it is you that is unkowledgeable, uninformed, and inexperienced with AI and I know this because one of my companies is among the most advanced in the world in AI research much of which exceeds my ability to fully understand, assess, and comprehend however I have an extraordinary team of engineers, mathemeticians, and physicists who are capable of explaining to me the basic, fundamental, core aspects of the science which is how I know you are just wrong. You can dismiss, reject, and disqualify outright any thing you want without providing any valid substantiation, proof, or reference however such is not science you are acting on "feeling", intuition, and "gut" which is fine of course for discussion but that will not withstand examination, study, and scrutiny by those actually working in the field of science.
Post removed 
Cassirer has been dead for 80 years. The things we know now, were not even conceptualized yet alone understood while he was alive.
Plato is dead for 2 millenias and is more advanced in many areas of knowledge than the main population including you... Try Archimedes and you will learn more than technology You will learn how to think by a man dead for 2 thousand years....Who gives a damn if he dont know how to train neural networks? He will learn that in less than one hour...Not a too complex mathematic...Grothendieck is a bit more difficult to chew try it...But you seems to idolizing technology...Not me... I prefer more sophisticated ideas and maths...

You keep getting hung up on (to me) neural networks as just another implementation of computing, with algorithmic repeatability, not as much as Andy2, but still guilty.

Still guilty? You are funny....

I even speaks about linking quantum computers to neural networks computers remember? And I speak about the differences between A.I. "creation" of concepts and the creation of concepts in human language...(various optimization mathematical methods for training neural networks are in no way an explanation for the intelligence of humans sorry)I does not advocate the superiority of human at all... My point is about human language.... My dog and any living system is conscious like you and me ...But language is a powerful tool to restructuring the brain individually and collectively...This is my point...

But in language the fundamental fact is the linkage between the motivated biological body and the gradually non motivated abstract symbolism and syntax... Then concepts creation in human is linked to an" in the world" and "out of the world "simultaneous process that is also a basis fact of language... The language is the basic and more fundamental TOOL there is ,and there will be, + scripture, his basic recreation.... It is the reason why I refer to Cassirer one of the more underestimated philosopher of science of this century...Read about him before trashing him for techological ignorance...

And, as far as I can tell, you start from the assumption humans are "special", or maybe biological creatures in general?,

Inform yourself and you will learn something about the social characteristic of living system (not human only and mainly except for language) and the consciousness of all living system....

Patronizing about neural networks, do you even have basic knowledge in linguistic?

Grothendieck expertise was mathematic (pure mainly), and frankly has no chops w.r.t. AI, and certainly not modern AI.

You dont even know that Grothendieck is one of the most profound thinker (yes in the last 2 centuries maths was NOT about simple calculus more about concept creations ) of the history of Maths...You think it will takes more than 5 minutes for him to learn optimizing maths behind neural networks? Grothendieck wrote thousand of pages about concept creations by the way....In french…. :)

We are all ignorant but I idolise knowledge.... I dont arrogantly push for technology against philosophy or spirituality, or worst against true mathematical science....Technology is not science....Only a small part of science.... Archimedes was a thinker with a method not a prehistorical Edison....

You can correct me about details of technology linked to dac, or amplifier, or any other tech artefact.... I bet you are engineer after all... But your general understanding is less than many here, because you are in a tunneling vision like any tech fad or specialist... Human thinking is a bit more vast and complex than you even imagine....The mathematics behind neural networks are in no way difficult and advanced... The neural network training is an advanced idea in technology yes.... Not the maths behind it...Then explaining the superiority of a neural network A. I. over human is the proof you dont understand anything more than the basic tech...