Cut the blind testing crap - Page 3  

 

It still doesn't change anything. And to bring this full circle, the measurements are reality, not perception. Even if I perceive the stick to be smaller, its not. Its really that simple. 
 

 

So you are saying you would happily use a small branch to prop up a bridge, because it measured to the needed length?
 

Measurements are parameterized models of reality, not reality. The words we use to describe perception are a kind of model of reality, not reality. You need to study inferential statistics.

 

You are veering way off the point. The point is that to know if you have an adequate model, you need to know what you are using the item for.

 

 

Feel free to direct me to an inner-ear microphone and equipment that can measure every aspect of sound, because AFAIK nothing of the sort exists and probably won't for several more decades.

 

 

In that case, I completely agree with you.

This is still wrong. even though measurements are used to parameterize models, they are not themselves the model. You can argue that the model is mistaken, or poorly applied, but that doesn't change the fact that the wavelength of yellow light measures 570–590 nm or that the standard concert A measure 440 Hz.

I have studied inferential stats and done both stochastic and deterministic community ecological models, btw. The main influence a model has on measurements is that what you think is true about a system will lead you to measure somethings and not others. You can be very wrong about what you've chosen as an accurate or complete model. And if you're arguing that people are looking at the wrong thing when considering audio equipment, fine. But then leave off about measurement itself being flawed.

However, in engineering, an end-point analysis is done before the product design is even considered.  For the most part, this is going to go in a couple of different directions depending on the audio preferences involved (tubes vs. ss amps, for example).  However, you would be able to define the parameters for the end-point analysis, saying things like, "I want to design an amp with <0.01% THD because DBTs have determined that it is less than the audible threshold."  At that point the measurement for THD & THD + N becomes a validation of your analyzed end point.

 

Fact is, non-linear and linear distortion make up 99% of what we hear, so performing a THD+N plot and an IMD plot will tell you 99% about the distortion products of the sound.  FR graphs are idealized, however, within the context with which they are built, they tell you a great deal about how a person could perceive the relative levels of the sound.  (Probably better would be to take a page from Nyquist/Lyapunov criterian and perform time-invariate analysis in the S-plane, but that's a different story).

 

In short, we know a great deal about what we need to do to make a product sound good based off of the measurements, because we can logically define those endpoints (based on earlier studies performed).  We actually know a great deal about how/what humans perceive in a sound wave.

 

 

Of course not! Those are definitions, not measurements.

 

Measurement is not "flawed", it's just not reality.

 

People often look at measurements out of context. You have to look at the whole picture to see my point.

 

You say that concert A measures as 440. Once I wrote some software to measure the frequency of a sine wave. It's all models. It's about fitting a model, and in the presence of noise. Maybe the pitch you are measuring is fluctuating, but your answer has to be a single number. Clearly you can see your measurement is a parameterized model, not the full richness of reality. Maybe the harmonics don't line up exactly. As soon as you say, "I measured the pitch as 440" you are talking about your model fit and you have fundamentally shifted from reality to a model.

 

I'm not saying that's bad. But then you have to ask what you are using this model for. Why do you care what the pitch is? Until you answer that, there is no way to say if your model is adequate.

 

 

 

 

 

If I measure something and it fluctuates, then that has to be part of how I report the measurements. It would be something like " I recorded X for Y minutes at Z intervals. Over that time period the average X was Q units, with a standard deviation of P."

See, not hard at all. There's no reason a measurement is done once or to think there's a single answer. I don't know why you keep insisting about the measurement being a model fit. Take the concert A for example, the definition of that has changed over time. You care what the pitch is so that people play in tune. That's not a model of reality, that's a decision about how to set a consistent reference point. It's 440 now, but it's been different things in the past. That doesn't mean that if you could time travel and measure it, that the earlier ones are wrong, just a difference in the local consensus.

Your exercise of writing a program to measure the frequency of a sine wave is specifically about a model, but I don't get why you don't understand that all measurements aren't.

What is this immeasurable reality you're referring to? Are we having a philosophical difference of opinion? (Which is where these things seem to end up in every thread.)

Reporting the answer as "average Q" and "standard deviation P" is a model. You use inferential statistics to compute those numbers, and the math involved is a model fit.

 

Measuring the wavelength of one cycle is a model too, because noise, bias, and measurement error will be present. It's not as simple as checking the zero crossings and thinking you've pinned down reality.

 

You don't HAVE to report the fluctuations. It depends on how you plan to USE the measurement. All measurements are in a context---the context of how they are used. A measurement outside a use-context is meaningless.

 

EDIT: I accidentally created some confusing when I said "Maybe you have to give a single number"-- I don't mean you always have to give a single number-- but sometimes you do. For instance, if you are writing software to run a kind of tuner device that musicians use to check their pitch, and the device has a single display, you have to give a single number. If you are programming a police-radar-speed-gun thingy, you have to give a single number.

 

 

I didn't say it's hard. I said that measurements are parameterized models.

 

If you made the effort to write a program to measure something, anything at all, I guarantee you that you will find out that you are fitting a model.

 

Edited by mike1127 - 8/22/11 at 10:07am