A how to for a PC XO.

[JoshK]

Originally posted by ewildgoose



Doing multiple sweeps at different points in a room is also an interesting idea, but it seems extremely difficult (to me) to figure out the maths of averaging the results? Presumably one could time align them based on group delay, but what would it mean to do a simple average? You would need some equivalent of a complex-average, ie taking into acount both phase and freq response? I haven't found many research papers on how you might do this, but I have done a bit of searching.

Personally I think an interesting idea would be to look at variation around the average and reduce the IR towards zero in places where there is lots of variation and less so where you have strong agreeemnt. The idea would be to leave an IR which shows the main reverb and reflections which are strongly present in several locations, but not show (and hence not correct) for the reverb which appears out of phase in different places. Just a thought, but I haven't tried to implement it yet...


Ed W

Your thoughts are a lot similar to mine. I haven't done any real research on this front yet but my brainstorming led me down a similar path. I vaguely remember a course I had as an undergrad in Response Surface Methodology (Stat course). Anyway, the idea I had was to examine the surfaces created by looking at plots in different directions to see where the problems really exist and disregard more of the variation do to combing or minor movements. I'd imagine you'd find more room problems and ringing through this "filtering" of your response. I had hoped that you could derive a smarter filter as a result.

[ewildgoose]

Yes, my idea was to take the IR, align the group delay approximately using the largest peak (say). Then simply look at the correlation of each point in each IR. Points with high degrees of similarity would tend to move to the average, those which were wildly differing would have to move towards some kind of interpolation

I don't think it would work too well in practice though because you would need so many IRs to get any kind of statistical significance... I guess that you could use a rolling window along the IR in order to increase the amount of data...

Another idea might be to stick to simple freq response for the multiposition averaging, and only use phase information for the primary listening position.

All the ideas simply boil down to reducing the strength of the correction at certain freqs/time, so guess another way of approaching the whole thing is to build a strong filter, then reduce the effect at certain freqs after the event...

It's tricky anyway...

Good luck

[JoshK]

Originally posted by ewildgoose Another idea might be to stick to simple freq response for the multiposition averaging, and only use phase information for the primary listening position.

This was my idea. Examining FR at multipositions to gain a better understanding for what you really want to correct for and what you don't. Phase for one position, although the same analysis could be done I would imagine for phase too. Kinda like partial derivatives, hold some variables constant and then analyze in one dimension at a time. Phase would be trickier to analyze so it would be revision 0.26 or so.

[ewildgoose]

Intuitively I don't really see how you can average (explicitly) the phase information?

Consider an intuitive example for a moment. Imagine speaker and listening position on the center line of a "box". The direct sound is exactly that, and after a delay you get the reverb from a side wall. Now measure at the point the sound would be reflecting off the sidewall - the direct sound will arrive in half the time, and the reflected sound will arrive instantaneously.

How do you reconcile those two IRs? Frequency is indirectly affected as a result of cancellation. Min phase is of course going to be broadly the same in both locations (although on and off axis is of course not identical), but the excess phase is totally different and of course uncorrectable in general.

It's a tricky bugger isn't it

Ed W

[fcserei]

I know it sounds like trolling here, but I don't get room correction (any more), and the posts about measurement just reinforce my wiews.
Having been a room correction advocate for a long time finally I gave up and revised my views.

IMHO, there is no correct way to "correct" a room, there is no such thing as room modes in music reproduction, you can't just take a room response and convolve it with a more or less inverse filter etc. What you get is different, subjectively maybe better, but not closer to optimal.

What you hear when you try to reproduce a musical event in a room is a superposition of the original signal reproduced by the speakers, and the ambient response of the listening room. The ambient response is the definite pattern of early reflections and later reverbs with Rt60 time up to a second in regular listening rooms.

The real problem with it is not that it will "color" the sound, but it will tell your brain everything about your listening environment - like how big is your room, what shape is it, how live is it, how far rare the speakers etc. (No wonder you will never confuse the sound of the hi-fi with the original - except some rare cases when the acoustical signature of your room accidentally closely matches the original recording venue.) This info coming not just from the amplitude decay of the orininal signal in the reverberant environment, but also from the directional/phase/amplitude/freq. response of the individual early reflections, and the direction/amplitude/freq. response distribution of the reverb.

When you do a single omnidirectional mic measurement in the room you are already doing a very crude averaging, keeping only the amplitude/phase decay info with MLS type measurements, and even worse, just the long term average amplitude with the sine sweep. Room modes itself are no more than long term amplitude averages of stationary signals in the echoic environment. This measurement is no way representative to the added sound (ambience!) of your room, so inverting it and using for room correction sounds useless. You will try to launch the complete averaged room ambient response form a single point together with the original signal, and hope for that will correct the full sphere ambient response. This is one of the reason behind the wiew, that you should use room corerection in the LF, where the amplitude avereaging shows huge FR anomalies, and the ear is not too sensitive for the directional info.

The only "working" room correction I can imagine nowadays would be a kind of ambisonic room response measurement/inversion for eight or more speakers with recursive algorhitms to correct the secondary effect of the correction signals, but ambisonics has it's own problem in the HF. At the end it just does not worth it. Why don't you just use heavily damped room and ambience synthesis.

PS. speaker correction is a completely different, and very useful topic.

[wigginjs]

Could someone please clearly define the difference between "speaker correction" and "room correction"? Are these two things actually seperate when you are correcting speakers that have been placed inside of a room? What good does correcting the amplitude/phase response of speakers in an anechoic environment do if you're just going to place them in a room and muck everything up again?

[ShinOBIWAN]

Originally posted by m0tion
Could someone please clearly define the difference between "speaker correction" and "room correction"? Are these two things actually seperate when you are correcting speakers that have been placed inside of a room? What good does correcting the amplitude/phase response of speakers in an anechoic environment do if you're just going to place them in a room and muck everything up again?

Ambiguous at best.

Your guess is as good as mine and I expect each has a different definition.

I think of them as:

Speaker correction is XO treatments to flat the anechoic response of phase, amplitude etc.

Room correction is physical treatments or using EQ to alleviate room problems.

[fcserei]

Speaker correction is an attempt to make the speaker as linear as possible in phase and FR in ideal (anechoic) environment.

The room acoustic itself is a linear (although not minimal phase) phenomenon, so if your speaker has a bump at a freq. your total room response will also reflect it, even though an averaged measurement could shows a dip at that freq.

[ShinOBIWAN]

Originally posted by fcserei
I know it sounds like trolling here, but I don't get room correction (any more), and the posts about measurement just reinforce my wiews.
Having been a room correction advocate for a long time finally I gave up and revised my views.

IMHO, there is no correct way to "correct" a room, there is no such thing as room modes in music reproduction, you can't just take a room response and convolve it with a more or less inverse filter etc. What you get is different, subjectively maybe better, but not closer to optimal.

What you hear when you try to reproduce a musical event in a room is a superposition of the original signal reproduced by the speakers, and the ambient response of the listening room. The ambient response is the definite pattern of early reflections and later reverbs with Rt60 time up to a second in regular listening rooms.

The real problem with it is not that it will "color" the sound, but it will tell your brain everything about your listening environment - like how big is your room, what shape is it, how live is it, how far rare the speakers etc. (No wonder you will never confuse the sound of the hi-fi with the original - except some rare cases when the acoustical signature of your room accidentally closely matches the original recording venue.) This info coming not just from the amplitude decay of the orininal signal in the reverberant environment, but also from the directional/phase/amplitude/freq. response of the individual early reflections, and the direction/amplitude/freq. response distribution of the reverb.

When you do a single omnidirectional mic measurement in the room you are already doing a very crude averaging, keeping only the amplitude/phase decay info with MLS type measurements, and even worse, just the long term average amplitude with the sine sweep. Room modes itself are no more than long term amplitude averages of stationary signals in the echoic environment. This measurement is no way representative to the added sound (ambience!) of your room, so inverting it and using for room correction sounds useless. You will try to launch the complete averaged room ambient response form a single point together with the original signal, and hope for that will correct the full sphere ambient response. This is one of the reason behind the wiew, that you should use room corerection in the LF, where the amplitude avereaging shows huge FR anomalies, and the ear is not too sensitive for the directional info.

The only "working" room correction I can imagine nowadays would be a kind of ambisonic room response measurement/inversion for eight or more speakers with recursive algorhitms to correct the secondary effect of the correction signals, but ambisonics has it's own problem in the HF. At the end it just does not worth it. Why don't you just use heavily damped room and ambience synthesis.

PS. speaker correction is a completely different, and very useful topic.

Thanks for a counterpoint to all the gushing thats passed so far.

Like everything its completely personal. Accurate it may be but I agree that the potential is there to rob musicality if over done.

One of the reason why I correct only for amplitude rather than try and blanket everything. I think phase compensation can bring benefits but tailor the speaker to the room and get competancy into the design it isn't so much a problem. Physical treatments along with minimal EQ seem to give me the best results for my room and speakers. Others may prefer nothing or even every form of EQ under the sun to try to defeat the room.

Do it too much and you can hear all the details but everything sounds flat, boring and dull.

[ewildgoose]

Try DRC and see if you think it's unneccessary...

Room Correction is trying to remove the impact of the room you are listening in and hence you end up hearing more of the original reverb. It is kind of a superset of speaker correction which just limits itself to trying to correct a speaker to be as "flat" as possible (or whatever shape you want).

The problem with only correcting the speaker is that what you hear is room + speaker, hence you need to correct the end result which arrives at your ears rather just than the sound leaving the speaker (although at least having that decent is a good start)

You can't make a crap speaker into a perfect one using digital techniques. Nor can you perfectly fix a room. If you want REALLY high end audio then you need to start with some REALLY good speakers and have a really expertly designed accoustically correct listening room. However, even then DRC will help, but for the rest of the time we don't tend to have access to the above and so DRC is a cheap way to get us perhaps 40-60% of the way there while still living in a room that doesn't have egg boxes stuck to the wall...


Room Correction is Impossible (a reply from Denis Sbragion)
http://www.duffroomcorrection.com/wi...rection_limits

Some actually results of DRC in a real room (showing massive improvement in room response)
http://drc-fir.sourceforge.net/doc/drc.html#htoc197

What is room correction?
http://en.wikipedia.org/wiki/Digital_room_correction