How do you measure sound stage?

if you read perceptual encoder literature they have explored what has to be preserved in the encoding process for stereo - but mostly judged on loudspeaker systems

likewise Smyth has to know something applicable to make the SVS Realizer work - but head angle tracking is only a 1-d independent variable

any height/depth information is inferred by the brain from its expectations/experiences of real world sound - in a stereo recording there simply isn't 3-d sound field information 

The soundstage is not determined by frequency response, but mostly by phase differences between left and right ears, and reflections caused by the headphones themselfs. The frequency response could do something to the perception of HRTF's but really, it's unknown territory. This subject hasn't been researched much, in fact Mr Linkwitz was saying in a recent presentation that, although everyone places the loudspeakers in a room to optimise bass response, no one has actually studied the effects of wall reflections on imaging.

And to answer your question, you cant really measure the soundstage. You could measure the "echoes" but you won't really know what it means.

One of those waterfall graphs could tell you, i would guess.

I reckon soundstage is way to subtle to measure directly.  In my experience, however, soundstage is best with headphones that are extremely phase coherent --- meaning that the high- and low-frequency components are getting through to your ears without any time shift from differences in propagation delay.

Or sumpin' like that.

Also, soundstage is most strongly related to the quality of the recording. Some have it ... most don't.

Sound stage is a function of how left-right channels differences are perceived, so I don't see how it could be measured using the normal metrics of audio fidelity. The cheapest POS speakers in an untreated room could give an amazing sound stage if appropriate source music is played. So I'd say sound stage is mostly about how a recording is mixed, and what reverb and echo effects are used and how they're panned left-right. Yes, some speakers and rooms can reduce the apparent width due to reflections that drown out the more subtle cues in the music. But the sound stage itself is entirely in the source. At least that's my take on it.

--Ethan

Interaural time and level differences contain information about the location of a sound source (from the difference in time that a common source impacts the left and right ears) which the brain is very sensitive to.  Other cues like frequency dampening and echo delay could convey distance information.  As Tyll mentions, preserving this information would require very good recordings and headphones with very good phase fidelity.

Quote:

Originally Posted by XXII 

To Ethan and silverxxx: It's not just left/right phase coherence though, is it? If you muted the right channel, you would still get some sort of soundstage difference between a HD800 and a RS-1 (say).
 

Yes and No. Reverb added to a mono recording can give a sense of front-back depth. But that's not the same as sound stage which (to me) implies phantom left-right placement. And it's still a function of the source recording, rather than the gear or speakers etc. So in that sense it's not an audio parameter that can be measured with absolute values. That is, it's still a subjective experience that exists outside the gear.

--Ethan

The difficult thing is that there is no way to measure it except by reference to a recording.  Someone would have to come up with some sort of demo with a noise moving around in space, possibly with all sorts of background noises going off in an attempt to muck up the driver.  After that you'd have to put headphones on a person, use some laser eye tracking, and tell them to look at where it sounds like the noise is coming from or something.  Multiple trials and averaged results would be required.  I don't think there's any practical way to do it.

It's easy enough to make a synthetic soundstage on headphones using HRTFs.  Any half decent headphone will give you some sort of soundstaging if the signal has it in there to begin with.  The problem is that most music is mixed for speakers, and what gives you a soundstage with speakers usually doesn't with headphones.  This necessitates the use of all kinds of black magic in order to get speaker like soundstage from headphones, via recordings mixed for speakers.

Conceivably you could use the same sort of demo track I mentioned above and compare the recording of headphones playing on the on the dummy head to the recording of the dummy head sitting in an anechoic chamber in front of some reference quality speakers.  Microphones sensitive enough to determine pick up all that data will probably be astronomically expensive, and I doubt run to run consistency would be high either.  It would also possible to compare a headphone recording from a dummy head to to the output of an HRTF function and compare those.  On the downside, you can't really have a perfect function to compare the recording to.  On the plus side you can keep all the data and reapply any new and improved HRTFs to it in the future.  I'm assuming that the microphones would still be cost prohibitive.

Quote:

Originally Posted by maverickronin 

 

It's easy enough to make a synthetic soundstage on headphones using HRTFs.

Make a binaural recording using a binaural recording head.  No need to get synthetic about it.

If you can find all the music, movies, and TV you like in binaural format, then sure its better.  There isn't a whole lot out there though.

Quote:

Originally Posted by XXII 

I don't think it's as complicated you make it. You should be able to deduce something about soundstage from the analog signal that leaves (say your amp) and the analog signal that enters your ear (or a dummy head's ear as mentioned by a microphone). This should just be some parameter of the HRTF, no?

It complicated because other typical measurements have an easy baseline reference.  The source signal.  You play the source signal over the equipment and and measure the differences between the recording and the original.

The soundstage, as a mess of phase and frequency shifts with time delays and frequency dependent attenuation is an inherent part of the signal.  First you have to measure the soundstage in the signal and them you have to measure what the equipment does to it.

The first issue it what do we measure the signal against?  This is hard because we don't even know what we're aiming for.  Are we trying to make a stereo signal mixed for speakers sound 'live'.  Or binaural recordings?  Or a synthesized sound for a game or movie?  These are linked but not identical.  If we decide on the most common case, music mixed for stereo speakers, we still have more decisions.  Are we going to measure against real speakers, a separate binaural recoding of the same live event, or a mathematical model?  And will that model be of speakers or of a person watching the performance?

After doing all that you have to interpret the actual data.  Are we going to use straight percentages like we do with noise?  I think the soundstaging will get lost in the noise of all the other distortion.  This could be minimized by using single tones at different points in space, but being less demanding it would likely exaggerate the score of poorer headphones as compared to better ones.  With sufficient dedication the details could be worked out though.  I think it is possible that the components of soundstage can be teased from the mess of noise and the test recording can be compared to the reference recording to give us some kind of percent difference.  At the end there's one more surprise though.  What if a headphone has more soundstage than the reference.  You'll need positive and negative scores!  Zero would be the reference recording and I'll let the flame warriors work out which way is positive and which way is negative.

It can in theory be done, but its very, very messy.  I think this is about as far as us amateurs can get.  Any further involves rather nasty math and stupidly expensive equipment.  Maybe someone has very clever and easy way to do this though.  I'm sure as hell not perfect, but I have actually spent some time thinking about this before.

many drivers(like the cd3k) distort the phase like hell to send sound cues in different directions to the human brain...measuring the SS accurately is next to impossible IMHO, as it's also dependent on the ear position within the ear pad...one millimeter off and you're not in the sweet spot anymore.

Quote:

Originally Posted by XXII 

 

Wow, it's great to have your comments Tyll! Your observations are very interesting. How do you measure phase coherence? 

 

When all the spectral components line up properly, you get nice clean edges.  It's easiest to see in the squarewave response.

As you can see, the Denon (blue) stops ringing sooner and has a flatter top than the AT (red) indicating better coherence.