Measure from the source to the output?

I'm not sure exactly what you're saying here.

The spectrum analyser shows how much of what frequencies are present in the music at any given time. It has nothing to do with the headphones being used. Even if there's no transducer connected to a system, the spectrum analyser will do its thing.

Well sure, that's basically what a frequency response graph displays, except in a much more clear way, since the baseline for a response graph is a uniform volume across all frequencies, rather than the fluctuating volumes and frequencies of actual music.

You mean why do people disagree regarding what headphones sound like what?

Frequency response graphs show how a particular transducer outputs each and every frequency given the same input level. You can pretty easily tell if the headphones are bass-heavy, warm, balanced, or bright.

Take this graph, for instance:




The Sennheiser HD650, HD595, and HD555 are being compared. They are all relatively close in their frequency response, but will sound quite different from each other if you listen to each of them.

Frequency response graphs won't tell you anything about the soundstage, tightness, dynamics, or clarity. This is where most of the debate happens. While pretty much everybody can agree that headphone X has a lot of bass, some will feel that it has too much, while others feel that they need more. Some might feel that the bass needs to be tightened up, others might feel that it's great just the way it is. There really isn't a good way of measuring these characteristics, since it's mostly subjective at this point.

It also depends on what type of music you listen to. Somebody wanting cans that are great for classical music will desire something completely different from somebody who listens to rap.

Not only that, but everybody's ears perceive sound in different ways. As it is, we don't even hear sound exactly the way it really exists around us. Human ears have a natural spike around 3,000 Hz, meaning we perceive sounds at this frequency as being much louder than at other frequencies, even if they're really being played at the same sound level. We also have a roll-off at very low and very high frequencies. But again, everybody's ears will be different.

I must disagree with this. Looking at a headphone's frequency response, you can easily infer most of those qualities. For example, if a headphone's frequency response goes very high, say -3db at 41kHz, you will know that both tightness and clarity. Tightness, because a transducer capable of moving so quickly will be able to respond very rapidly to changes in the signal, and clarity because it will be able to transduce more information in a given time period than something with a more narrow frequency response.

As for dynamics, that is a quality of signals, not headphones.

I think we all agree that's true, but not meaningful in totality unless all we listened to were sine waves.

Music is a much more complicated waveform, and you cannot infer the true HRTF from the freq response curve. You need little microphones in the ears, just as Smyth uses and other posters here noted.

Then of course we all "hear" something different anyway, both due to the physical nature of our ears, and our brains.

Carefully designed, listener-blind testing analyzed correctly (which means not just averaging everyone's answers) will give us more knowledge. Use engineering measurements to separate out the awful phones from the good ones, then listener tests to finish the job.

But well-designed panel tests are hard (very hard) and measurements are easy if you have the gear. It is human nature that we do what is easy.

Note added: well of course for one particular short music passage you can take the FFT, then apply the freq response curve to each revealed freq, and get some idea of the HRTF at that moment, but you can't make a general conclusion about music ... except some vague genre-related comments perhaps. You also have to measure transient response. My main point is: listening tests are a direct frontal attack on the problem we are trying to solve.

You have to bear in mind that "tightness" doesn't mean anything. A lot of the terms used to describe sound used here are totally meaningless.

Indeed. A better and measurable term would be rise time or slew rate - the time taken to raise a signal from 10% to 90%. This can be used to characterize amplifiers, cartridges and CD players alike, though in the case of red book CD players the time is mathemetically deterministic viz related to the 1/44,100 thing..

For an amp varous figures are bandied about and the result depends on the power of the amp but most modern 50wpc amps will comfortably do 35v/microsecond which is more than adequate for 20 - 20,000 full scale output.

With speakers the picture is less clear cut as not only do you have to get the cone to respond but you have questions about decay when the signal drops.

The headroom frequency response graph may be precise for a frequency "sweep", but how accurate is it when the driver cone is trying to reproduce 12 different frequencies with varying amplitude, attack, decay, etc.

You have a point -- IMD can only be measured with multiple simultaneous tones.

A frequency response (or impulse response) graph can only completely characterize a linear, shift-invariant system. From The Scientist and Engineer's Guide to Digital Signal Processing (http://www.dspguide.com/): If headphones, amps, and DACs all met this description, there'd be no point in choosing any piece of equipment over another -- simply choose any combination, measure its impulse response (or frequency response -- they're two sides of the same coin, and from one you can get the other), and turn that into an "undo" filter.

Frequency response graphs do give useful information -- if you like a lot of bass, and you see that 80Hz is 10dB quieter than 1kHz on a headphone, you know you probably want something else. However, if the driver rattles every time you hit 80Hz, you won't know that from a frequency response graph!

Sweep signals are fairly standard for measuring frequency response. It's pretty accurate, although I don't personally possess data to support this.

White and pink noise are also commonly used. Here, the transducer is reproducing multiple frequencies at once.

I've taken courses in vibrations, dynamic system control, and acoustics in college. I have a fair amount of experience with similar topics. I don't claim to be an expert, however.

Frequency response - Wikipedia, the free encyclopedia

Edit: Gah, someone beat me to it with a better answer. Like I said, I'm not an expert!

IMD is the most important factor, once the pure frequency response has been established. The dynamics is what makes the difference between "good" cans (those with adequate frequency response). Transducers with a good dynamic response will produce clearly separated, well defined instruments, it's what most people look for.
So far most audio reviews i have read only consider the static frequency response. For dynamic aspects most tests are subjective... It would be interesting to see more scientific measurements of dynamics in audio equipment reviews.