DDF, thanks for the links to the ITU paper - very interesting. I have read the Griesinger paper before (and his collection, which is a great read). I first heard of David's work indirectly by using a Lexicon digital reverb in the studio. It was based on David's work on IACC, decorrelated reverb, etc. Very realistic sounding and many recordings, including surround sound now use his algorithms: DAVID GRIESINGER (LEXICON): Creating Reverb Algorithms For Surround Sound
If you liked the Haas effect, this demo on interaural time difference may be interesting: https://www.youtube.com/watch?v=CuYNFv2Oc08 Another aspect to why early reflections will mess with localization from a music reproduction perspective in listening rooms.
It is also one of the reasons why one will see control rooms with the dichotomy of not having parallel surfaces, yet the left and right side of the control room are mirror images of each other. There is a measure for that called inter-aural coherence coefficient which is a measure of channel and room reflection equality for the first 80 milliseconds of sound travel. I know of only one commercial software package that can measure this. Anyway, another topic area for another time.
Like you say not taking away from other's preferences of spaciousness versus pinpoint accuracy, my preference is for the latter.
marco is on the money. Typically 20 to 25 dB SPL is the dynamic gap and dare I say, is also indicated in the ITU, EBU, Dolby guidelines linked earlier that the monitoring needs to be capable playing through the range of 80 dB SPL to peaks of 105 dB SPL at the listening position with no dynamic range compression. Also see Bob Katz's article on calibrated monitoring linked to earlier.
One may also find the Dynamic Range Database useful for finding out which pressings of your favorite recordings have the most dynamic range: Album list - Dynamic Range Database Note that DR is only one attribute, albeit an important one, to determine the quality for any given recording.
While on opposite sides of this world, in different rooms with totally different speakers here's a side by side comparison of Mitch's favoured room curve compared to what I ended up liking the most in my room:
The curves look very similar to me. I didn't start with his curve as a target, I grew to like this over time.
I do have to say our room treatments do have strong similarities.
Just to comment on this, I also did not start with this curve. I used the B&K to start with and iterated through dozens of small variations and is interesting to me that wesayso and I have ended up with very similar target response.
Looking at the latest Harman/Toole paper on house curves also shows an almost identical preferred response to what wesayso and I have come up with independently. Other's that are employing house curves also show similar responses. I don't feel it is coincidence.
While I agree that building or buying speakers with a flat anechoic response and wide dispersion is great, the reality is that the moment any speaker is placed into a room, the tonal response/balance as heard at the listening positon is going to change due to room effects.
So much so, when I was still working in the recording industry and spoiled rotten listening to music in acoustically designed control rooms, costing several hundred thousand dollars, I gave up on having a home stereo as it sounded like a poor facsimile of what I heard at the studio. Had little to do with the speakers or electronics, but all about room acoustics. Unfortunately, fully designed and acoustically treated (not dead) rooms are fairly expensive.
Flash forward a number of years where software DSP and computers became sophisticated and powerful, we now have software DSP that can greatly assist in dialing in ones preferred tonal balance and counteract some room effects. And for me, maximizing my gear investment/potential and listening pleasure.
Going back to a control room example where one side is a mirror image of the other, and where the listening position and distances to the monitors have been laser measured down to a 1/4" tolerance or less. Unless one has a room like that, then one is going to hear and measure a different frequency response between the left and right speakers. This audibly affects the imaging. Check out that interaural time difference video linked to earlier.
So using DSP to simply balance the frequency response so that the amplitude from the left and right speaker is as close to equal (ideally < +- 1dB variation between them) over the frequency range at the listening position will audibly enhance imaging quality and phantom center image. Just my preference.
Personally, I feel that there are a half dozen or so acoustic guidelines/specifications (linked in this thread
that one can measure at the listening position which can quantify an accurate music reproduction system. Again, if that happens to be ones preference. Will post more on this at another time.
hi mitch, I dont know why, but the curve you and weysayo use is a bit too much in my room. Ive settle with flat to 2khz, then straight line to -3db at 10khz, -4db at 20khz.
I cannot agree more with the effect of treatment in a room.
very weird that diyaudio do not include a room acoustic section where people can share their method to treat the room/ building panels/bass traps.
Its incredible the difference bass traps makes for LF performance.
I cannot agree more with the effect of treatment in a room.
very weird that diyaudio do not include a room acoustic section where people can share their method to treat the room/ building panels/bass traps.
Its incredible the difference bass traps makes for LF performance.
Last edited:
Feels like people compare apples & pears sometimes when it comes to flat/accurate versus various house/target curves.
All well having a flat standard for anechoic on-axis response I guess (although such speakers will also vary widely in perceived tonal balance due to different dispersion/directivity/ power response) but any attempt at establishing a standard (or even reaching a consensus for the best) in-room response must first lay down the rules for how this should be measured. The response at the listening position was not meant be equalised flat (unless all reflections are windowed out, which in itself will be quite useless in terms of low-end extension and resolution, save perhaps for tweeters). Seems some still believe this is how it should be done ("accurate, like they do in the studios").
As soon as any amount of reflections are taken into account, as in let into the measurement, the response will start displaying a fall in the treble and a boost in the bass due to the behaviour of sound and acoustics of a "small" listening room. But the slope of this tilt will - and should - vary with the length (and to some extent the type) of the window. A target curve for a steady-state RTA response should not be the same as that for a shorter impulse response (and this is where I think some of the cinemas in Toole's latest paper got it a bit wrong). Mimicing the BK or similar curve only makes sense if it also mimics the intended gating. Tracking it exactly to the half decibel in say Audiolense won't by itself translate very well.
Even these improved "psychoacoustic" representations of FR, as implemented in REW, Holmimpulse, Acourate, Audiolense etc, are only comparable if they use exactly the same assumptions for the "dynamic" window (i.e. time as a function of frequency). Once you do, then you get "surprisingly" similar results or consensus among such trained listeners as Mitch and wesayso - who I assume have used the same software (Acourate?) to measure their respective in-room response (and I'm guessing the difference in the top-end is due to either hearing (e.g. age) or more likely how much their respective tweeters beam, where different toe-in can matter too)...
Looking at many target or actual in-room responses in this thread ("I prefer this response after much testing..." - "Really, I find it too much in my room"), very few are clearly labelled how they have been or are meant to be measured.
It seems to me that a method similar to what Harman is using, with a defined listening window, is a good starting point spatially - but I appreciate that it might be rejected for being too complex as it requires averaging of multiple measurements (and a consensus on the right "weighting" of direct and indirect sound). See here for an explanation of this approach:
Speaker directivity / off axis response: theory and measurement techniques - Acoustic Frontiers
Next step should be to decide on the gated window to use, and why not a dynamic one at that.
Then, and only then, can we argue what such a curve ideally should look like and compare measurements
All well having a flat standard for anechoic on-axis response I guess (although such speakers will also vary widely in perceived tonal balance due to different dispersion/directivity/ power response) but any attempt at establishing a standard (or even reaching a consensus for the best) in-room response must first lay down the rules for how this should be measured. The response at the listening position was not meant be equalised flat (unless all reflections are windowed out, which in itself will be quite useless in terms of low-end extension and resolution, save perhaps for tweeters). Seems some still believe this is how it should be done ("accurate, like they do in the studios").
As soon as any amount of reflections are taken into account, as in let into the measurement, the response will start displaying a fall in the treble and a boost in the bass due to the behaviour of sound and acoustics of a "small" listening room. But the slope of this tilt will - and should - vary with the length (and to some extent the type) of the window. A target curve for a steady-state RTA response should not be the same as that for a shorter impulse response (and this is where I think some of the cinemas in Toole's latest paper got it a bit wrong). Mimicing the BK or similar curve only makes sense if it also mimics the intended gating. Tracking it exactly to the half decibel in say Audiolense won't by itself translate very well.
Even these improved "psychoacoustic" representations of FR, as implemented in REW, Holmimpulse, Acourate, Audiolense etc, are only comparable if they use exactly the same assumptions for the "dynamic" window (i.e. time as a function of frequency). Once you do, then you get "surprisingly" similar results or consensus among such trained listeners as Mitch and wesayso - who I assume have used the same software (Acourate?) to measure their respective in-room response (and I'm guessing the difference in the top-end is due to either hearing (e.g. age) or more likely how much their respective tweeters beam, where different toe-in can matter too)...
Looking at many target or actual in-room responses in this thread ("I prefer this response after much testing..." - "Really, I find it too much in my room"), very few are clearly labelled how they have been or are meant to be measured.
It seems to me that a method similar to what Harman is using, with a defined listening window, is a good starting point spatially - but I appreciate that it might be rejected for being too complex as it requires averaging of multiple measurements (and a consensus on the right "weighting" of direct and indirect sound). See here for an explanation of this approach:
Speaker directivity / off axis response: theory and measurement techniques - Acoustic Frontiers
Next step should be to decide on the gated window to use, and why not a dynamic one at that.
Then, and only then, can we argue what such a curve ideally should look like and compare measurements
Last edited:
I know Mitch uses Acourate to create his FIR correction. I've used his online chats with Bob Katz from the time they both used Audiolense as sort of a guide to get myself started. We've also exchanged a few PM's talking about this type of signal manipulation.
I should probably also mention member gmad as a great help to get me going with using DRC-FIR, my choice of poison. He has started a thread in the full range forum offering help to anyone willing to dive into that sort of processing: http://www.diyaudio.com/forums/full-range/275730-convolution-based-alternative-electrical-loudspeaker-correction-networks.html
I use DRC-FIR with my own target curves to get my desired results. I also use heavily modified templates/settings and window lengths (varying with frequency in number of cycles) based on my own extensive testing. I'm not using the standard set as found in the DRC-FIR setup.
Both graphs from Mitch and myself are measurements from REW using the same settings.
For me to rely on measuring at one point and to start using FIR correction I first did a test; measuring a FIR corrected speaker at several locations around my listening area (a big couch) and average that result. The average of those measurements was nearly identical to the desired curve at the listening spot. I probably need to redo that test as the last time I did that I had less acoustical treatment than I have now. But it gave me the confidence to use a tool like DRC-FIR and measure at one single sweet spot. It is true my tweeters fall short at the high end (In this graph it is exaggerated by the 1/3 smoothing used) and they extend to about 17 kHz (Measured about 10 degrees off axis). I have no real tweeters, woofers or subs, all of it is produced by an array of 25 3.5" full range speakers.
To arrive at the curve I use now I didn't specifically follow any other curve. I grew into this curve over time by making small adjustments. Sometimes resetting it entirely and restarting this process. That's all I can think of right now to explain a bit of my background arriving at this particular curve. I hope this gives some insight.
I do believe there is quite a bit of difference between using FIR correction (with linear phase corrections) and using IIR EQ. I tried both but invested way more time in optimising the FIR correction as to me it seems like the most powerful tool to use for my specific application. I rely on EQ to get a listenable result with my arrays. That's very different from starting with a speaker with flat response in an anechoic room.
I should probably also mention member gmad as a great help to get me going with using DRC-FIR, my choice of poison. He has started a thread in the full range forum offering help to anyone willing to dive into that sort of processing: http://www.diyaudio.com/forums/full-range/275730-convolution-based-alternative-electrical-loudspeaker-correction-networks.html
I use DRC-FIR with my own target curves to get my desired results. I also use heavily modified templates/settings and window lengths (varying with frequency in number of cycles) based on my own extensive testing. I'm not using the standard set as found in the DRC-FIR setup.
Both graphs from Mitch and myself are measurements from REW using the same settings.
For me to rely on measuring at one point and to start using FIR correction I first did a test; measuring a FIR corrected speaker at several locations around my listening area (a big couch) and average that result. The average of those measurements was nearly identical to the desired curve at the listening spot. I probably need to redo that test as the last time I did that I had less acoustical treatment than I have now. But it gave me the confidence to use a tool like DRC-FIR and measure at one single sweet spot. It is true my tweeters fall short at the high end (In this graph it is exaggerated by the 1/3 smoothing used) and they extend to about 17 kHz (Measured about 10 degrees off axis). I have no real tweeters, woofers or subs, all of it is produced by an array of 25 3.5" full range speakers.
To arrive at the curve I use now I didn't specifically follow any other curve. I grew into this curve over time by making small adjustments. Sometimes resetting it entirely and restarting this process. That's all I can think of right now to explain a bit of my background arriving at this particular curve. I hope this gives some insight.
I do believe there is quite a bit of difference between using FIR correction (with linear phase corrections) and using IIR EQ. I tried both but invested way more time in optimising the FIR correction as to me it seems like the most powerful tool to use for my specific application. I rely on EQ to get a listenable result with my arrays. That's very different from starting with a speaker with flat response in an anechoic room.
Last edited:
mitch, InOtIn beat me to the punch. What window do you apply to your room curve? I imagine the B&K curve was with an RTA given how old it was (ie very long ave time). Harmon provided no clear guidance but used many window sizes in their paper. I've tried MLSSA's variable Adaptive window but couldn't find a sweet spot graphically that reflected what I thought I heard.
If I can ask one more question, regarding dynamic headroom, what frequency dependence do you find in recording? Once had to spec a fast compressor for telecom audio and after analyzing numerous Cds and reading the literature, found that peak values were consistent across frequency almost up to 10kHz (IIRC). Only difference is dwell time of the peaks was much shorter higher up in frequency you went. Any guidance from the recording world on peak/ave margin necessary vs frequency? And if it was lowered at high frequencies, was this compensated with different attack/release times? I always felt uncomfortable with the recommendation to use lower powered amps on tweeters when biamping with passive xovers (active can take advantage of the tweeter's greater sensitivity, but that's different)
Regarding room symmetry and close tolerances, they make a big difference. I designed and had built an IEC room with low NC, high STC, but knowing it was an "audio room", the construction lead took it on himself to make walls on every dimension perfectly parallel to the mm, definitely not on request. If looking for a theoretical display of perfect flutter echo, this was it. Took many $$$ to add diffusion and absorption to fix. Real eye opener on what half an inch will do to room acoustics.
If I can ask one more question, regarding dynamic headroom, what frequency dependence do you find in recording? Once had to spec a fast compressor for telecom audio and after analyzing numerous Cds and reading the literature, found that peak values were consistent across frequency almost up to 10kHz (IIRC). Only difference is dwell time of the peaks was much shorter higher up in frequency you went. Any guidance from the recording world on peak/ave margin necessary vs frequency? And if it was lowered at high frequencies, was this compensated with different attack/release times? I always felt uncomfortable with the recommendation to use lower powered amps on tweeters when biamping with passive xovers (active can take advantage of the tweeter's greater sensitivity, but that's different)
Regarding room symmetry and close tolerances, they make a big difference. I designed and had built an IEC room with low NC, high STC, but knowing it was an "audio room", the construction lead took it on himself to make walls on every dimension perfectly parallel to the mm, definitely not on request. If looking for a theoretical display of perfect flutter echo, this was it. Took many $$$ to add diffusion and absorption to fix. Real eye opener on what half an inch will do to room acoustics.
Happy to have holes poked in the idea. Here it is, what do you think?I must have missed it. I am always glad to point out where you are wrong.
http://www.diyaudio.com/forums/mult...t-vs-accurate-hi-fidelity-14.html#post4482749
Been there, done that... and yes, found one after spending countless hours tweaking cycles at hi-mid-low frequencies while observing the shape of the IR in the filter procedure designer in Audiolense...
... but this is beside my point, which was about a standard or common foundation for comparing in-room measurements and target curves, and not about finding my own personal preference.
My idea about the sum of on and off ear frequency response didn't get much traction, I see.
I asked why binaural dummy heads weren't used more in evaluating the in-room response here and Earl commented that he had experimented with this quite a lot actually. Such a technique should capture what you are driving at I guess.
I think that it is a good idea, but one that would need to be developed into a real procedure.
I only did this as a recording for playback not as a means of taking measurements. I couldn't get the recordings to sound right so I am not sure that using this same technique for measurements would be meaningful.
cool, what is your preferred curve?
the standard in pro forum is to measure the room at the listening position isnt it?
about standard to compare in room measurements, I believe this is pointless as we all have very different room especially among audiophile. Room curves preference have been quite a lot covered by Harman, toole, ect.
Last edited:
Yes, we had also discussed size and shape of the dummy head. Earl mentioned bowling balls, IIRC.
Therein lies the problem. It needs to be general enough to be applicable to most situations, E.G., a certain size sphere or egg shape with 2 mics on it. And there are things to think about, like does the off axis mic need to be delayed before it's summed? Where should the measurement be taken, etc.
I find it easier to pick a house curve that works well in the room, and go with that. Mostly the house curves differ in the hinge point.
With respect to in-room measurement analysis and correction, JJ Johnston has a presentation on the Acoustic and Psychoacoustic issues in Room Correction. The first half of the presentation up to slide 31 I feel represents the best psychoacoustic knowledge of small rooms we have wrt what our ears care about at the listening position. From a room analysis perspective, jj is talking about frequency dependent windowing (FDW).
Most acoustic measurement software today has some facility for measuring using FDW. Some software have both FDW for analysis and correction. Most typically follow jj’s recommendations where the window is large at low frequencies (e.g. 500 ms) and narrowing to almost direct sound at high frequencies (a few ms or less). A visual of what that looks like can be found at http://drc-fir.sourceforge.net/doc/drc.html#htoc29 I find the default FDW settings in most of these software packages work just fine.
In my room, I have measured speakers with narrowing directivity as frequency rises, and as I have mentioned elsewhere in the thread, I for one am not a fan as it locks your head in one position at the listening position. I have since moved to a speaker system that uses constant directivity waveguides. However, the target response I used was the same for both and produced the same tonal balance I prefer. The only difference I found was with constant directivity devices is the sweet spot covers the entire couch area instead of the head in vice only spot at the listening position with narrowing directivity speakers. Others have reported similar results with a wide variety of speaker polar responses. If one thinks about it a bit, using FDW, with a short window (i.e. virtually direct sound) at high frequencies, it makes sense that the target response is going to be the same with little effect from the room due to the very short window.
The reason for the differences in the high frequency extremes between wesyaso’s and my measurements is that I have a tweeter and he does not. Got nothing to with our ears or our rooms. What I am pointing out is that wesyaso’s speakers are very different than mine (line array with no XO versus old school 3-way Klipsch Cornwall’s in a sealed cab with mid and high frequency constant directivity waveguides. Bob Katz speakers and room is also very different from wesayso’s and mine using 2 satellites and 2 subs, but also similar target response. All of us are using different measurement and correction software, but with a common theme of FDW. While some of us tinker with the FDW settings, I found it relatively insensitive as long as the majority of the impulse response is within the window. If using FDW, and understanding how it works, then it would appear the selection of house curve is based on one’s tonal preference more than any other factor.
DDF, I sent you a PM about the compressor.
Most acoustic measurement software today has some facility for measuring using FDW. Some software have both FDW for analysis and correction. Most typically follow jj’s recommendations where the window is large at low frequencies (e.g. 500 ms) and narrowing to almost direct sound at high frequencies (a few ms or less). A visual of what that looks like can be found at http://drc-fir.sourceforge.net/doc/drc.html#htoc29 I find the default FDW settings in most of these software packages work just fine.
In my room, I have measured speakers with narrowing directivity as frequency rises, and as I have mentioned elsewhere in the thread, I for one am not a fan as it locks your head in one position at the listening position. I have since moved to a speaker system that uses constant directivity waveguides. However, the target response I used was the same for both and produced the same tonal balance I prefer. The only difference I found was with constant directivity devices is the sweet spot covers the entire couch area instead of the head in vice only spot at the listening position with narrowing directivity speakers. Others have reported similar results with a wide variety of speaker polar responses. If one thinks about it a bit, using FDW, with a short window (i.e. virtually direct sound) at high frequencies, it makes sense that the target response is going to be the same with little effect from the room due to the very short window.
The reason for the differences in the high frequency extremes between wesyaso’s and my measurements is that I have a tweeter and he does not. Got nothing to with our ears or our rooms. What I am pointing out is that wesyaso’s speakers are very different than mine (line array with no XO versus old school 3-way Klipsch Cornwall’s in a sealed cab with mid and high frequency constant directivity waveguides. Bob Katz speakers and room is also very different from wesayso’s and mine using 2 satellites and 2 subs, but also similar target response. All of us are using different measurement and correction software, but with a common theme of FDW. While some of us tinker with the FDW settings, I found it relatively insensitive as long as the majority of the impulse response is within the window. If using FDW, and understanding how it works, then it would appear the selection of house curve is based on one’s tonal preference more than any other factor.
DDF, I sent you a PM about the compressor.
Yes, we had also discussed size and shape of the dummy head. Earl mentioned bowling balls, IIRC.
Therein lies the problem. It needs to be general enough to be applicable to most situations, E.G., a certain size sphere or egg shape with 2 mics on it. And there are things to think about, like does the off axis mic need to be delayed before it's summed? Where should the measurement be taken, etc.
I find it easier to pick a house curve that works well in the room, and go with that. Mostly the house curves differ in the hinge point.
Pano, I am having good success with a pair of these mics: The Free Space - Binaural Microphone - $499.00 : 3Dio, Free Space Binaural Microphone - Binaural mic for binaural recording Jeff, the owner, has other mics, including where one can just purchase the ears and DIY your own setup. I feel he has done a good job if you listen to his sample binaural recordings. I know he has tweaked those ear shapes to give the best sound for a wide variety of HRTF's.
A few hurdles, one is being able to calibrate the mics. I have a few ideas on how to achieve that, but have not got around to it yet. Measurement take at the listening position
- Status
- This old topic is closed. If you want to reopen this topic, contact a moderator using the "Report Post" button.