Cut the blind testing crap - Page 2  

Personally, I have no faith in any attempt to objectively describe transducers. There are so many factors at play, an objective measurement or a verbal description is always going to be inferior to simply sitting down and listening to it to make your own decision. I'd never buy speakers without auditioning them with a selection of alternatives first.

 

Regarding measuring sound and measurements, I think you misunderstand the problem. We can produce many precise measurements, but that doesn't mean we've completely described the behavior of the headphone. Try out this statement:

 

"There is nothing in human personality that can't be measured."

 

I think everyone would agree that human behavior is not reducible to measurements. That's also true of audio equipment for the same reason, although this may not be obvious. Think about it.

 

Think about this statement:

 

"An audio device is a behavior."

 

Think about what that means for the usefulness of measurements.

 

 

 

What does personality have to do with audio? It will only create preferences, which are unrelated to objective performance.

 

"There is nothing in human perception that can't be measured."

 

I think that's a true statement. The goal of objectively measuring audio equipment isn't to measure preferences, but measure accuracy. We have a strong grasp of what's physically perceptible to humans. We also have a strong grasp of how electricity works, how sound waves propagate, and how a driver moves and distorts. We can't account for placebo, and the test the OP proposes isn't trying to.

 

How about instead of me thinking about "An audio device is a behavior", how about you explain what that means and why that's relevant?

 

The OP's test would capture exactly what the headphone produces to the limit of the testing environment and microphone. There's no hidden magic property we hear that a microphone can't. Any differences in perception between our ears and a microphone is caused by the limits of our hearing, or by our brains altering our perception. That doesn't alter the accuracy of the headphone, just our preferences. What the microphone captures is what the headphone produces. That aspect of the OP's test would work fine. The question after that isn't "So which is more pleasurable?", but "Which is most accurate by these standards?" I thought that was the goal of "high fidelity"?

 

Let me address two points.

 

POINT 1

You could claim that we can "measure" the behavior of a person-- after all, we can take video of their movements and map them out precisely. We can get audio of the sounds they make with their mouth-- heck, use the most accurate microphone you can get, and the best digital recording.

 

This is of little use because it only shows us the behavior of the person in a limited number of situations.

 

It doesn't help us predict the behavior in all situations.

 

This is exactly analogous to audio. Measurements are made with a few inputs. This does not help us predict behavior to all inputs.

 

One would only think that measurements could predict all behavior if one thinks mathematical models are perfect descriptors of a device.

 

POINT 2

 

Perceived accuracy is subjective. Say we listen to a recording of a singer. What you think sounds most like a live singer probably differs from I think sounds most like a live singer because we focus on different aspects of the signal.

 

POINT 1

 

Headphones don't have personalities. They don't just decide on a whim to produce a recording a different way than they did the last time. The physics behind them is considerably less complicated than the human brain. They're fed a signal, and the driver moves accordingly to produce sound waves. Those sound waves can bounce around, interfere with other waves, and leak out. It's predictable, though less predictable than an amp or DAC.

 

The measurements by the OP would be done with real music. They will be indicative of real world performance. Why would they only apply to the songs tested? The headphone doesn't care if it's being fed Bach or Lady Gaga. If it can accurately produce one, it can accurately produce the other. Why, specifically, would that not be the case?

 

POINT 2

 

I already said I'm not talking about perceived accuracy. I'm talking about objective accuracy. Similarity to the original source, after accounting for different types of field equalizations. The goal is not to make something sound subjectively good. The goal is to make it sound like the original recording. Perception doesn't even come into play. You're welcome to prefer a headphone that doesn't measure up. People love Grados, even I do. Their measurements are not pretty, and I wouldn't call them accurate.

(1) The purpose of an audio system is to be listened to. So only perceived accuracy is relevant.

 

(2) Models don't fully describe reality. Measurements are a model.

 

1. Not when it comes to objective performance. If only perceived accuracy mattered, why review products? No two people will hear the same thing.

 

2. You're welcome to criticize the OP's test methodology with specific variables not being accounted for, like I have. That would help his measurements better model reality.

1. Yes, it complicates matters that only perceived accuracy is relevant, but you can't just pretend models equal realty and make the problem go away.

 

2. I don't need to provide any specifics in order to claim that models are not reality. Every scientist knows that models are not reality.

 

 

 

So we shouldn't model anything because it will never be reality?

1. Probably true

2, Not true, unless you're using some interesting definitions of the words measurement and model.

Models aren't reality - but measurements are.  They are the sum of reality and measurement error.

 

That doesn't mean measurements are a complete assessment of reality - they only assess what they measure.  If you measure the length of an object, you don't know its width unless you measure that too.  That doesn't make the measured length of the object wrong.

 

It is important to distinguish between quantities that have a measure and what can be measured - just because we don't have the means of measuring it with our equipment doesn't mean it isn't possible, nor that it doesn't have a measure.  Obviously we're not going to get around Heisenberg's uncertainty principle, but both quantities which we can't measure at the same time are ultimately defined as quantities.  We know that an electron has a given position and a given momentun, although we can't know what they both are at the same time.  We know that our brains are made out of neurons that send electrical impulses to each other (and obviously a lot more than that), but we can't measure or model everything yet.  That doesn't mean it is impossible, unlike for the HUP.  Unless the brain/human thoughts consist of more than just quantities with a measure - i.e. the supernatural - then it may well be possible to measure and model human brains and thought completely some day in the future.

 

 

 

 

Anyway, while I wouldn't say such a project is useless, we have the distinct problem that music is very never mastered with the intent to be listened with zero added distortions of any kind.  There's not an objective standard for what recordings should sound like in the first place.  It's an artistic venture.  Recordings are mastered to sound good on speakers - with room reflections perhaps intended to be a part of the sound.  Headphones, car stereos, etc. are all considered for most recordings.   The objective standard is to accurately reproduce what the artists/engineers/producers hear (or want to hear) on the recording with minimal distortion, not to reproduce what is specifically on the recording with minimal distortion.

 

Binaural recordings get us as close as we can right now to that, I would say.  I still like loudspeakers for normal recordings - probably because they actually bring me objectively, measurably closer to what recording artists (i.e. musicians, producers, engineers, etc. - whomever has the final say) intend the music to sound like than headphones that reproduce the recording itself with lower distortion.

 

Now, if we could just get a neural measurement of what recording artists hear when they tweak everything to their liking.  Or perhaps, we could attempt to make ambiphonic-like maps of sound fields and reproduce that as close to reality as possible.

I'm addressing the claim that "we can measure everything." That is false. Models can be useful if you understand what you are doing with them and their limitations. These concepts come from a branch of mathematics called inferential statistics. Measurements are parameterizations of models. Let me give an example.

 

Take a frequency response graph. I don't know if this is going to be clear to you without studying linear systems theory, but a frequency response graph is a model of how a linear system behaves. Well, a headphone is not a linear system. That's why the model is not reality. An FR graph fits numbers to the model. That's why it's called a parameterization.

 

Furthermore, unless you think FR graphs are pretty and worth framing, their only purpose is to predict the behavior of the headphone. And really, if you look at the whole context, the actual goal is to create an aesthetic experience in a listener, so in the biggest picture the point of an FR graphic is to predict a human experience.

 

You can say anything you like about how flat graphs are good, etc. As soon as you say that, you have lost the context.

 

The usefulness of a model has to be tested against how it is experienced. A lot of people don't like that because it's a "sticky problem," no longer nice and objectively safe.

 

 

Measurements are not reality. They are parameterizations of a model.

 

When you measure the length of a stick, you probably don't realize this, but you choose a model. Do you model it as an object that doesn't change its length under changing temperature, humidity, etc? That's probably the simplest model. But unless you like to masturbate to numbers, there is always a higher purpose. Are you measuring this stick because you want to use it to prop up something? Do you model it as incompressible? Do you model it as compressible and use a linear spring model? If that's good enough for your purposes, sure.

 

The trick in audio is to understand the purpose of the model, which is to predict a human experience.

 

 

 

To say it won't make a difference functionality, you have to know what you are going to use the stick for.