|
Post by flak on Dec 13, 2014 17:16:09 GMT -5
The common frequency response graphs show the frequency response in one single point in space... this literally means one point in space where the mic is placed, but measurements are different in different positions... even a few centimeters make a difference up to the point that what we measure at one ear is different from the other one. So we want to correct the common behaviour of the different curves from different measurement positions as we do not listen with one ear only in a rigidly fixed single point in space Here we see a forumer's eighteen curves of the nine measurements for left and right channels in the nine points that define the listening area together with the two curves (the lighter ones) that represent the averages of the respective L & R channels. The individual curves are shown in the full Dirac Live version for PC/MACs only, the XMC-1 shows the most practically useful and meaningful ones, the two lighter ones in this image... this is the behaviour relative to the two channels BEFORE correction. This is what happens AFTER Dirac Live's correction... (some may wonder how is it possible that the averages are the ones shown in these images seeing that the peaks are only a few dBs high while the dips are much lower, those are logarithmic scales and 70 dBs plus 70 dBs add to 73 dBs and not 140) One may wonder how reliable are Dirac Live's graphs when compared with measurements from third party dedicated applications... Michael Lowe, an indipendent forumer, has taken the time to verify that by using Fuzzmeasure on his Mac and a reference Earthworks M23 microphone... but the same can be done with REW on a PC, REW has all the necessary features. This is his image of the two channels as shown by Dirac Live BEFORE and AFTER correction (the kind of image you see in the XMC-1): This instead is his image of the eighteen measurements (both channels combined, 1/24 octave smoothing) by Fuzzmeasure BEFORE Dirac Live correction: while this is the image of the 18 measurements by Fuzzmeasure AFTER Dirac Live correction: By comparing them we can see that the Dirac Live's graphs are quite accurate. You can read the full thread here: www.whatsbestforum.com/showthread.php?15212-Dirac-validationOne last remark... the same considerations apply in the time domain so the impulse response also varies with position but impulse response evaluation is too long to discuss here, may be in another post Ciao, Flavio
|
|
hemster
Global Moderator
Particle Manufacturer
...still listening... still watching
Posts: 51,920
|
Post by hemster on Dec 13, 2014 18:01:03 GMT -5
|
|
|
Post by nickwin on Dec 13, 2014 18:03:37 GMT -5
Nice post, thanks for sharing that. You can see from the pictures you posted that when everything is working properly the real room response is very close to Dirac's estimation. What about when things don't go perfectly? Why not just do one more sweep of measurements of the post EQ response to visually confirm that the sound your ears are actually getting in the room is really what it is suppose to be? I've seen a couple people post here wondering if the measurements or upload were corrupted somehow, a final measurment would erase all doubt.
|
|
|
Post by socketman on Dec 13, 2014 18:27:52 GMT -5
Flak .....Having you come to the forum and share your knowledge and insight is greatly appreciated. Thank You
|
|
Lonnie
Emo Staff
admin
Pay no attention to the man behind the curtain
Posts: 6,999
|
Post by Lonnie on Dec 13, 2014 18:53:03 GMT -5
Nice post, thanks for sharing that. You can see from the pictures you posted that when everything is working properly the real room response is very close to Dirac's estimation. What about when things don't go perfectly? Why not just do one more sweep of measurements of the post EQ response to visually confirm that the sound your ears are actually getting in the room is really what it is suppose to be? I've seen a couple people post here wondering if the measurements or upload were corrupted somehow, a final measurment would erase all doubt. When the Dirac software is uploading into the XMC-1, their is a system in the code that verifies complete and accurate uploads. If a packet is lost or corrupted, an error message will be shown, but if the XMC shows complete on the front panel, then that means it all went fine. Lonnie
|
|
|
Post by nickwin on Dec 13, 2014 19:10:20 GMT -5
Nice post, thanks for sharing that. You can see from the pictures you posted that when everything is working properly the real room response is very close to Dirac's estimation. What about when things don't go perfectly? Why not just do one more sweep of measurements of the post EQ response to visually confirm that the sound your ears are actually getting in the room is really what it is suppose to be? I've seen a couple people post here wondering if the measurements or upload were corrupted somehow, a final measurment would erase all doubt. When the Dirac software is uploading into the XMC-1, their is a system in the code that verifies complete and accurate uploads. If a packet is lost or corrupted, an error message will be shown, but if the XMC shows complete on the front panel, then that means it all went fine. Lonnie Thanks Lonnie. That eliminates the possibility of a bad upload but what about bad measurments, even if it's user error? There are some minor differences between Dirac's estimation and the measured results in the op with the Dirac graph being slightly more linear than the measures one. Is that level of difference about average? How far off can it be in the extreme cases? I'm always going to wonder how close I am to that green line. There's something to be said for seeing real measured results with your own eyes. Yes, I'm nitpicking. Even with the above concerns it looks you guys have created a real winner with the XMC1. Congrats! NOTE: I want to clarify that I am not implying that Dirac could or should correct for user error or major speaker flaws. I was thinking more along the lines of having a visual cue that those things are present in the form of a final measurment.
|
|
klinemj
Emo VIPs
Honorary Emofest Scribe
Posts: 14,744
|
Post by klinemj on Dec 13, 2014 19:20:01 GMT -5
If you are really concerned, check it with another software...like REW.
Mark
|
|
Lonnie
Emo Staff
admin
Pay no attention to the man behind the curtain
Posts: 6,999
|
Post by Lonnie on Dec 13, 2014 19:36:21 GMT -5
When the Dirac software is uploading into the XMC-1, their is a system in the code that verifies complete and accurate uploads. If a packet is lost or corrupted, an error message will be shown, but if the XMC shows complete on the front panel, then that means it all went fine. Lonnie Thanks Lonnie. That eliminates the possibility of a bad upload but what about bad measurments, even if it's user error? There are some minor differences between Dirac's estimation and the measured results in the op with the Dirac graph being slightly more linear than the measures one. Is that level of difference about average? How far off can it be in the extreme cases? I'm always going to wonder how close I am to that green line. There's something to be said for seeing real measured results with your own eyes. Yes, I'm nitpicking. Even with the above concerns it looks you guys have created a real winner with the XMC1. Congrats! A couple of things to consider here. First off, nothing, not even Dirac can fix user errors. The whole process is pretty simple and straight forward, but it has no way to know if the end user screws up. All it can do is work with what it is given. Second, speakers are mechanical devices that eherantly are non-linear in their function. Thus, all Dirac can do is correct the electrical responses to the drivers. Non-linearities of the transducers themselves will always be present and the main difference between the graph Dirac shows and what was measured has more to do with the sampling rate and type of smoothing applied. Transducers typically range in non-linear response from 10 to 30%. So as far as I'm concerned Dirac is dead on the money. Even with the best EQs out there, I seriously doubt anyone could get it as close as Dirac has, but that is just my opinion. Hope this helps. Lonnie
|
|
|
Post by nickwin on Dec 13, 2014 19:40:55 GMT -5
Well all I can say is I can't wait to get it in my room
|
|
Lonnie
Emo Staff
admin
Pay no attention to the man behind the curtain
Posts: 6,999
|
Post by Lonnie on Dec 13, 2014 19:42:15 GMT -5
Well all I can say is I can't wait to get it in my room I think you will be very pleased and if you are concerned about making a mistake in the test, please feel free to give us a call. We would be happy to help. Lonnie
|
|
|
Post by nickwin on Dec 13, 2014 19:48:26 GMT -5
If you are really concerned, check it with another software...like REW. Mark I guess if wanting to be sure I'm getting the most out of my hardware and software is being concerned then yes I am as most of us are. I'm no saying it's a huge deal or anything it would just be nice. Maybe I'm more concerned than I need to be, as an Audyssy user im more than familiar with erratic results. Lonnie says Dirac gets within +\- 3db of target which is great and has pretty ended any doubts I had. As far as going in after and using REW that seems like a lot of extra work when I already have my computer running Dirac and the xmc1 mic already hooked up and in place and already have software running totally capable of measuring the room again and overlying the results.
|
|
|
Post by urwi on Dec 14, 2014 3:26:51 GMT -5
I agree but humans also don't listen to an average. And just because the average improved doesn't mean all points making up that average improved equally. Do you have measurements of single locations at and next to the mic locations that show how all of them improved? The common frequency response graphs show the frequency response in one single point in space... this literally means one point in space where the mic is placed, but measurements are different in different positions... even a few centimeters make a difference up to the point that what we measure at one ear is different from the other one. So we want to correct the common behaviour of the different curves from different measurement positions as we do not listen with one ear only in a rigidly fixed single point in space
|
|
|
Post by flak on Dec 14, 2014 10:07:01 GMT -5
Hi Urwi, the question I was trying to answer was the one about the accuracy of Dirac Live's "forecast" of the measured response in the room as mentioned in the title of the thread... we can see that those graphs are rather accurate, the individual curves instead (which of course you can validate) are not shown in the XMC-1 implementation (a wise choice) because Dirac Live is not intended as a measurement application while the average curve is the tool that is indispensable for the user to modify the target. As you know the same measured frequency response which is optimal in a room can sound worse in another room and/or with other speakers... so that's the reason to exist of the average curve as a tool (this also hints at the fact that we cannot rely on a frequency response alone to judge about listening results). So the answer about the accuracy of Dirac Live's graphs is a positive one but you raise another question... are those graphs, even if accurate, a real improvement? You suspect that while the average has improved the single individual measurements might have deteriorated instead. You may draw your own conclusions by looking at the nine individual graphs (if we assume they are accurate) at nine different points in the first images that I have posted by comparing them before and after correction, or you can use independent measurement applications if you think they are not accurate enough as well as measure at other additional different points if you think that those nine ones are not representative. In any case it seems to me that what you are saying is in accordance with what is explained in a document of ours... "A correction which is good for the mean response is very different from a correction which on average is good for any one of the measured responses" The same document also adds that there is more than that... that we should not mix up Fourier transforms with perceived frequency responses and that perceived frequency responses are time-dependent. It also states that a joint time and frequency analysis is required and that the problem of spatial variations must be considered carefully. That detailed document (that goes as far as possible in detailing what we can make public) is the following: diracdocs.com/on_room_correction.pdfFlavio
|
|
|
Post by lbrown105 on Dec 14, 2014 10:38:50 GMT -5
|
|
|
Post by urwi on Dec 14, 2014 10:45:07 GMT -5
Hi Flavio, The paper also says that both of the following graphs are representative of what we hear. So what does the averaged curve in the Dirac Live tool perceptually represent? What factors determine the correct curve? Speaker dispersion? Reverberation time? What performance parameters would one need to look at to determine the correct curve?
|
|
|
Post by flak on Dec 14, 2014 13:50:40 GMT -5
Hi Flavio, The paper also says that both of the following graphs are representative of what we hear. So what does the averaged curve in the Dirac Live tool perceptually represent? What factors determine the correct curve? Speaker dispersion? Reverberation time? What performance parameters would one need to look at to determine the correct curve? Hi Urwi, I recall now that after looking at those two images from our paper you asked a very good question.. "how Dirac Live handles those two different situations. How to optimize transient and stationary sounds at the same time?" You may remember that I answered that your question was too good to be answered that I preferred sidestepping it and that all I told you was that we basically base the frequency response correction and impulse response correction on very different data based on the nine measurements but pre-processed in different ways, looking at different time scales and averaging in different ways... nothing has changed since then. The averaged curve in Dirac Live is a tool to verify the correspondence with the target curve as they cannot always match... take also into account that the same exact measured frequency response curve can be "right" or "wrong" and this is determined by many factors as you correctly point out. Reflections represent a large part of the contribution to the measured frequency response and they are affected by the directivity index of the speakers as well as by the contents and dimensions of the listening room, the recordings also play a role because different recording techniques will feature different combinations of direct and reflected sound. As a result the "correct curve" that you mention unfortunately does not exist... if it existed we could use the same target curve in all instances and Dirac Live's option of modifying it would be useless Ciao, Flavio
|
|
|
Post by urwi on Dec 15, 2014 3:46:51 GMT -5
I don't agree that the recording should play a role in determining the correct room correction approach or target curve. The goal should be to reproduce whatever is on the recording as faithfully as possible. But let's not open up this can of worms So you're saying that somehow the correct curve is a result of how the speaker interacts with the room. So my question still would be, "How to determine the room curve for a specific room/speaker combination?" What's Dirac's stance on it? Reflections represent a large part of the contribution to the measured frequency response and they are affected by the directivity index of the speakers as well as by the contents and dimensions of the listening room, the recordings also play a role because different recording techniques will feature different combinations of direct and reflected sound. As a result the "correct curve" that you mention unfortunately does not exist... if it existed we could use the same target curve in all instances and Dirac Live's option of modifying it would be useless Ciao, Flavio
|
|
|
Post by flak on Dec 15, 2014 6:47:16 GMT -5
Dear Urwi, I don't want to open a can of worms but in my opinion we cannot disregard the real world fact that there are no standards for the recordings as summoned for example here: seanolive.blogspot.it/2009/10/audios-circle-of-confusion.htmlYou certainly remember the target curve suggested in this forty years old document: diracdocs.com/BruelKjaerSceltaDiffusori.pdfIt still has merits today but we should not overlook that it is a reasonable compromise which works well for many recordings but is not carved in stone... you will notice for example that they state that a flat target curve could be used instead if recordings were made under far field conditions with a suitable mixture of direct and reflected sound and a pair of ruler flat B&K measurement microphones. In a few words the Dirac Live target curve which can be easily modified as necessary seems very useful to me Ciao, Flavio
|
|
|
Post by urwi on Dec 15, 2014 14:12:45 GMT -5
This is certainly true for music but not so much for broadcast (see ITU standards) and movies (see SMPTE standards). So my question still would be, "How to determine the room curve for a specific room/speaker combination if the original recording is produced to a known standard?" What's Dirac's stance on it? in my opinion we cannot disregard the real world fact that there are no standards for the recordings
|
|
|
Post by flak on Dec 16, 2014 10:38:00 GMT -5
This is certainly true for music but not so much for broadcast (see ITU standards) and movies (see SMPTE standards). So my question still would be, "How to determine the room curve for a specific room/speaker combination if the original recording is produced to a known standard?" What's Dirac's stance on it? in my opinion we cannot disregard the real world fact that there are no standards for the recordings Unless I'm missing something (which is always possible) I don't see a problem... if you want to conform to a standard (i.e. SMPTE for a "small room" of 5.300 cubic feet) you can replicate those target curves if you want to. I'm not sure it would be appropriate in our normal listening rooms because of the differences with a theatre but it is feasible in Dirac Live... Ciao, Flavio
|
|