Pieface Posted July 17, 2015 Posted July 17, 2015 People talk about having a flat response in room. Our technical paper reading types assure us that under blind test circumstances flatter frequency response is preferred by listeners with normal hearing. Flatter frequency response is portrayed as desireable and necessary for HiFi reproduction. What we don't see is much data posted by our users of in room responses from systems that are generally considered to be pleasing or of high quality. This leaves a bit of a knowledge gap for people like me in terms of how to adjudge where the point of diminishing returns of chasing a flat frequency response starts to bite. At this point is what is a realistic expectation in a typical living room type scenario? Is it +/-3dB? Does it vary depending on the octave you are considering? At what level of octave smoothing are we judging the "flatness" of the response? Is flatness more important to our perception of fidelity than other loudspeaker properties? What do we consider acceptably flat when more than one listening position is considered? Maybe I waste too much time thinking about this stuff but i am ever curious. Anyone got 2 bobs-worth to throw in?
Satanica Posted July 17, 2015 Posted July 17, 2015 The flatter the better, don't overthink it. Human perception is within about 0.5db if I'm not mistaken.
andyr Posted July 17, 2015 Posted July 17, 2015 AIUI, flat response for the so-called audio band (20-20KHz) is not a good recipe - what we want (for sound you want to listen to) is for the treble to shelve down ... something to do with the Fletcher-Munsen curve, I believe. Andy 1
Satanica Posted July 17, 2015 Posted July 17, 2015 AIUI, flat response for the so-called audio band (20-20KHz) is not a good recipe - what we want (for sound you want to listen to) is for the treble to shelve down ... something to do with the Fletcher-Munsen curve, I believe. I think this is wrong, the Fletcher-Munsen curve is about hearing a flat frequency response irrespective of volume. 1
Pieface Posted July 17, 2015 Author Posted July 17, 2015 The flatter the better, don't overthink it. Human perception is within about 0.5db if I'm not mistaken. I'll just use 1/1 octave smoothing then. Great success! 2
3G_ Posted July 17, 2015 Posted July 17, 2015 An in-room flat response sounds very thin and bass shy with excessive treble. http://www.hometheatershack.com/forums/rew-forum/96-house-curve-what-why-you-need-how-do.html
aechmea Posted July 17, 2015 Posted July 17, 2015 You want Acoustical Measurement Standards for Stereo Listening Rooms Nyal Mellor, Acoustic Frontiers LLC & Jeff Hedback, HdAcoustics 2011 http://www.acousticfrontiers.com/wp-content/uploads/2011/10/acoustic_measurement_standards.pdf
Pieface Posted July 17, 2015 Author Posted July 17, 2015 Can we not get sidetracked with a target curve discussion? Pretend I said how closely your frequency response tracks your target curve or something if that makes life easier Aechmea. Awesome. Will read over the weekend. TY.
hochopeper Posted July 17, 2015 Posted July 17, 2015 (edited) Flat anechoic response is preferred. That isn't in room. Edited July 17, 2015 by hochopeper 1
3G_ Posted July 17, 2015 Posted July 17, 2015 The Mellor Hedback paper is one of the most succinct papers on the desired properties of a listening room. Well worth a read.
Guest Peter the Greek Posted July 17, 2015 Posted July 17, 2015 Its interesting, I've found I prefer a roll off at the top end, so long as its smooth through off axis. The way I found this out is I prefer the sound of my system sitting on the floor as opposed on on the sofa, so I measured it....what do you know it rolls off real nice from 12-20khz - not unexpected given the nature of the tweeter I use It'll be interesting to see how I react in a dedicated and appropriately treated room
davewantsmoore Posted July 17, 2015 Posted July 17, 2015 Anyone got 2 bobs-worth to throw in? Sure. There's a couple of aspects to consider..... but it all essentially boils down to how does what the chart is telling you, correlate to audibility. What do we mean by flat? There is literally flat.... where the level at 20hz is the same as 200hz, and 2khz, and 20khz... This is rarely what is literally meant.... because depending on how it's measured, this doesn't usually sound good What we are typically looking at, is no unintended peaks and dips. The width of the peak/dip being a key attribute of how it will sound.... with generally wider peaks and dips, being more audible than narrow ones. For example.... if we had an unintended dip averaging -3dB .... across the frequency range of 1000 to 4000hz. Then this speaker would sound spectacularly different, from one which didn't. At this point is what is a realistic expectation in a typical living room type scenario? Is it +/-3dB? +/-3dB can be used as a guide "guide". Worse than this is very(!) bad, and depending on the width and location of the peaks or dips..... extremely audible, as a change in balance, tone/timbre. +/- 0.5dB is my target ..... although some errors are better left uncorrected (and this is really a big catch IMHO) At what level of octave smoothing are we judging the "flatness" of the response? The best way, is to use no smoothing ..... but when we look at the unsmoothed chart, remember that small (narrow) wiggles are much less audible than wide changes..... Often ones which "look bad" don't sound bad, and vice versa. eg. a smooth looking 2dB dip spread over a couple of octaves could make a speaker sound much "duller" than one without..... where as violent looking up/down/up/down over a narrower frequency range, isn't as audible in practise. This leaves a bit of a knowledge gap for people like me in terms of how to adjudge where the point of diminishing returns of chasing a flat frequency response starts to bite. You can probably tell so far, that not all flat responses are created equal..... and assessing the response can get quite complicated..... the question of "does this wiggle in my response matter?" ..... really comes back to the question "what is causing it? Is flatness more important to our perception of fidelity than other loudspeaker properties? This is a really interesting piece of the puzzle... and nearly the most important one ..... and it's very instructive to ask the question "what other properties?". The thing is that there are very few other "properties" of a speaker, that are not already represented in the flatness of the response. In general, if you have a flat frequency response .... then these other properties (like the impulse response, phase, time alignment of drivers, etc.) .... are also perfect. What this means, is it is a mistake to prioritise "something else", over the response "flatness" ..... because they're just different views of the same thing. "the response". What do we consider acceptably flat when more than one listening position is considered? This is arguably the most important piece of the "response flatness" discussion. We often see people look at a frequency response chart..... and make an analysis and draw conclusions from that.... but there are an infinite number of frequency response charts (from every point in space). .... a good speaker has an equally flat response from lots of angles.... and this is where charts which show the response over a range of angles, either by showing an average, or by showing all the different angles, can very very enlightening. A practical example of this.... is that we could have two speakers ..... both measure perfectly flat when measured directly in front .... but when measured from 30deg to the side .... one is still flat ..... but the other has a big dip from 1000 to 4000 Hz. They will sound spectacularly different. Maybe I waste too much time thinking about this stuff but i am ever curious. It's complicated but very important. It's no surprise there are armies of people who don't "believe" that measurements correlate with system performance (audibility)..... it is very easy to take a measurement out of context, or make conclusions from it, which aren't supported when more data is considered. where the point of diminishing returns of chasing a flat frequency response starts to bite. Removing errors from the frequency response is THE path to improvement of performance. BUT.... for a person applying EQ to a speaker in-room..... there is the possibility that correcting the response to be flat(ter), can backfire. When correcting an error measured on one listening axis ..... we need to consider if the correction is a net benefit when considering other angles If you've got a big sharp dip in the speaker..... you can brute force it away with EQ ...... but now you may have a big peak when measured from other locations. 6
davewantsmoore Posted July 17, 2015 Posted July 17, 2015 (edited) . Edited July 17, 2015 by davewantsmoore
davewantsmoore Posted July 17, 2015 Posted July 17, 2015 (edited) . Edited July 17, 2015 by davewantsmoore
davewantsmoore Posted July 17, 2015 Posted July 17, 2015 (edited) Pretend I said how closely your frequency response tracks your target curve or something if that makes life easier Alrighty then. :-) +/- 0.5dB ..... and considering all angles of radiation. If the EQ doesn't work as a net benefit over many angles of radiation... then your speaker has problem which probably shouldn't be tacked with EQ. Edited July 17, 2015 by davewantsmoore
Pieface Posted July 17, 2015 Author Posted July 17, 2015 (edited) Thanks for your detailed response @@davewantsmoore Just for clarity when you say you are pursuing 0.5dB "flatness" and you are looking at unsmoothed data, do you mean you consider the data unsmoothed 20Hz-20kHz?... Are you looking at something that resembles the red line as it all "comes together" as you get closer to a "flat" response or are you looking at something that resembles the blue line but through experience/knowledge knowing which "jaggies" are irrelevant in terms of audibility? I keep learning with every one of these threads so that is a good thing Edited July 17, 2015 by Pieface
davewantsmoore Posted July 17, 2015 Posted July 17, 2015 Are you looking at something that resembles the red line as it all "comes together" as you get closer to a "flat" response or are you looking at something that resembles the blue line but through experience/knowledge knowing which "jaggies" are irrelevant in terms of audibility? Your picture is a great example. The ultimate goal would be to remove all the blue jaggies..... but most of them are caused by diffraction, and reflections from the room ..... and correcting these with EQ, is almost always ill advised. However the 1/3 octave response shows the trend where we have broad dips in the response. If we call 70=flat, then we have an roughly octave wide dip from 1-2khz and 4-8khz. These will be particularly audible..... BUT they need to be considered along with the response at other measurement angles. There's no point correcting these dips, if they don't exist at other listening angles..... or at least we might correct them differently based on other angles. In practise one would normally use a very small amount of smoothing. Say 1/24 or 1/48 octave ..... and then look at the curve keeping in mind that the big broad deviations are more important than the small wiggles.
murrmax Posted July 17, 2015 Posted July 17, 2015 Yes actually there's no such thing as a flat frequency response, ... you should look at fixing the room first before apply eq, trying to fix things with eq is suboptimal as mentioned above 2
Pieface Posted July 17, 2015 Author Posted July 17, 2015 Cheers for the input guys. I'm certainly not looking at implementing EQ as the first horse out of the barrier. I'm still plugging away with the theory side of things and trying to build the bridge of understanding how the theory relates to what I see in the measurements when I moves things around. For example I read a discussion here a couple of weeks back about radiation patterns from speakers and the timbre matching of reflections. Someone linked a talk by Toole IIRC. So I took a few measurments off axis this is the speaker at 1/12 smoothing: Now I have no idea if this is good, bad or indifferent but with what I have learned today hopefully I might be able to take some further measurements and try and gain some insight into why I prefer this speaker to an alternate speaker which appears to have a flatter response measured on axis: Cloth ears still the leading explanation :lol:
svenr Posted July 18, 2015 Posted July 18, 2015 AIUI, flat response for the so-called audio band (20-20KHz) is not a good recipe - what we want (for sound you want to listen to) is for the treble to shelve down ... something to do with the Fletcher-Munsen curve, I believe. Andy Not Fletcher-Munson but with the absorption and diffusion characteristics of the sound field. 1
davewantsmoore Posted July 18, 2015 Posted July 18, 2015 Cloth ears still the leading explanation :lol: Try making a chart for each speaker.... with 4 lines on each chart. 0 degrees (on axis) ... 15 degrees, 30 degrees, and 45 degrees (off to the side). Ensure that when you move the microphone off to the side, that it's distance from the speaker is the same (ie. move it in an arc). When measuring the two different speakers: Place them in the same location. Make that location as far from walls as possible. Put the mic about 2m from the speaker. Measure roughly the shortest path length of sound which will reflect to the microphone (so from the speaker, to the wall, to the mic). Yes. That's a lot of work Cloth ears still the leading explanation :lol: There's no accounting for taste Most people hear th phrase "flat response is preferred", and say.... "don't tell me what I like". Having a "flat response" is really about determining what errors the speaker has. When you send it a flat signal, and the output is not flat ...... the question "why" tells us a lot about the speaker performance. .... it doesn't mean that you can't set the response to whatever you like, and be happy.
svenr Posted July 19, 2015 Posted July 19, 2015 (edited) Flat anechoic response is preferred. That isn't in room. Well, it actually would be under two conditions: a) the speaker has constant directivity across the entire audio band (good luck designing one), and b ) the room has constant absorption and diffusion characteristics across the entire audio band (that's actually much easier to achieve than a) ). Edited July 19, 2015 by svenr 1
svenr Posted July 19, 2015 Posted July 19, 2015 (edited) I give you my 5 cents on that. First, you've gotta differentiate between two frequency regions in the room - the low frequency (LF) region where modal resonances dominate, and the high frequency (HF) region where statistical acoustics apply. Being divided around the Schroeder frequency, you might say this transition happens in around 150 Hz to 400 Hz depending on the overall reverberation time T60 in the high frequency region. When you do room response measurements at the listening spot, you've gotta bear in mind the great variance of the high frequency region. A little change of the mic position changes the response in that region drastically. So always use averaging over at least 5 spots with say 20cm spacing longitudinally and transversally. Then you must look at the two frequency regions independently. They've got nothing in common, so don't bother trying to draw analogies. In the LF region, look at the total response first (FFT over the entire measurement window). Using 1/24 octave smoothing to remove the highly variable noise, any remaining peak or dip will correspond to modal resonance effects and will be highly audible. A waterfall (with correct settings) can also demonstrate where resonances lie, but the waterfall puts too much emphasis on resonances in my opinion, and makes cancellations (dips) appear not so bad. So you better stick with the FFT. Now you should work it - move subs, apply EQs, get the time alignment between subs and mains right if the dip exists in the crossover region, etc pp. In the HF region things are a lot different. After measuring and 1/24 oct smoothing, you will see some line. The only general rule that can be stated is that it must be monotonic with minimal amount of undulations and should (if at all) decrease continuously with increasing frequency, but never increase. Any undulation that deviates from the smooth trend line by more than say 0.5dB will most likely affect the perceived sound quality. Whether it does cannot be said with certainty. Sounds all a bit vague and fishy? Well, it actually is vague and fishy. Dave already eluded on the core problem - the interaction between a speaker with frequency dependent directivity and power response, and a room with frequency dependent and also spatially dependent absorption and diffusion characteristics. Quite frankly, you will not be able to tell which of these elements - the speaker or the room - causes a certain undulation in the measured HF response. And here comes the important but about the HF region - you cannot fix room acoustic problems in that region with EQ. So if you find problems in the HF response, the first thing you should check is the reverberation time T60 in the HF region in 1/3 octave bands. If you find undulations there (for guidance see for example that little paper posted earlier), you may already be onto something. Think how you can possibly improve that mess. If the reverb time decreases greatly with frequency, that's the reason why your measured HF response at the listening spot decreases over frequency. More absorption or diffusion increases the rate of energy loss over time, hence there is less energy in the sound field. And since you are looking at a very long FFT window, you are capturing both the initial speaker response as well as the entire reverberate response. Makes sense, doesn't it? Very strong early reflections can also be the culprit in the case of strong undulation - check the ETC and find the reflection spots. Absorb the reflections with a big cushion and see if that has effects on the HF response and (most importantly) on the perceived sound quality. You can never have enough cushions at hand. And gaffer tape. And a camera to document the mess you made. And a notepad to write down which addition ("adult-sized kitty print cushion suspended 100mm below first reflection spot on ceiling") yielded which change to the perceived sound quality. If any of these additions makes a real difference, you should consider making the kitty cushion a permanent fixture. Ok, but what about those peaks and dips due to the speaker. They can be EQ'd, can't they? Well, I never attempt any EQ based on the HF response measured at the listening spot. The influence of the room is simply way too strong to get a meaningful EQ design together. If you are brave enough to interrogate your precious mains speaker, put it in the middle of the room as far away from the side walls as possible, place the mic at 1m distance from the midrange unit (only if you have conventional cone drivers, for large radiators like panels etc different rules apply), absorb the floor reflection with plenty of cushions, do the same on the ceiling if needed, and measure the response over +- 30deg both horizontally and vertically. Check the ETC after the first measurement for strong reflections and try to absorb them as well - the less reflections the more meaningful the whole exercise. Do some meaningful windowing with loosing too much frequency resolution (read some Joe D'Appolito if this doesn't make sense to you). Then average the measurements in both directions and view them together with the on-axis response (REW is very good for that). If you see big undulations in the HF regions and weird sheit above 3 kHz, call the guy you paid the money to and give him an earful. Such a scummbag, selling you that krapp. Bloudy heell. Then calm down. What you see in the averaged measurements you can carefully EQ (you might be able to do this in REW, apply EQ to each measurement that is. If not, write the guy who makes it and demand such function. Tell him you cannot work with the freeware crapp he publishes unless your request is acted upon immediately). Don't worry about the on-axis response, should this become a little forked up you can play with the toe-in to get it right. If you have a good quality measurement, flatten the response out as much as possible (and I mean flat as in horizontal). Be a bit careful if the on-axis and the averaged measurements diverge ever more at higher frequencies. This means your rubbish speaker has a completely poor directivity. Call that guy again and demand your money back. After that is done, apply the EQ in these divergence regions carefully. You can certainly overcompensate (i.e. create a peak) on-axis, but you may wanna stay a wee bit short of flat in these divergence regions. Otherwise you on-axis and near-on-axis sound may become coloured again which potentially cannot be fixed by toe-in. You can tell that this EQ exercise in cases of acute directivity pestilence may need some trial-and-error to get it right. Had you only bought speakers based on a real technical data sheet rather than that glowing review in this magazine which had such beautiful pictures of the speakers on the following advertisement page... Na well... Here is one tip - measure the second speaker under the exact same setup as the first one (on-axis and a few selected off-axis points). If there is deviation greater than +-0.5dB you have found yourself some wood for the next BBQ... In any case, use the same EQ settings for both sides of a stereo system. Always. Don't ask why. Just stick to it. So, you turned the room into a padded cell, fiddled with the mic and the EQ for hours, woke up the missus three times, got told to get stuffed and threatened with a messy divorce should the place still look like this in the morning. Ah, that sense of achievement. So what will it most likely sound like? Correcting the problems in the LF region as well as the time alignment between the sub and mains will give you a smooth and tight bass. By the way, push the bass pup by about 5dB somewhere below 90 Hz to get some real good feel. Be careful to not have that lift above 100Hz, otherwise everything will become very muddy. I sometimes use a low shelf filter plus an EQ to get the nice low end happening. Fixing the power response problem in the HF region should have been part of what you paid for. But in any case, your speaker now at least complies with some minimal quality standards for good sound reproduction. That will immediately make a better sound no matter what the room acoustics say. Fixing some nasty reflections will have stabilised the imaging and the apparent stage width. Try creating some very strong reflections from the side walls and see if you like that. Give a very realistic sense of involvement, especially for acoustic and classic music. Don't bother if you listen to umph-umph electronica. You may wanna pursue this further by improving the room in accordance with good acoustic engineering practice to yield acceptable variation of the reverberation time and a better diffusion characteristic throughout the room. But be careful before purchasing expensive panels; most things can be done by simple (and good looking) DIY. Edited July 19, 2015 by svenr 10
davewantsmoore Posted July 22, 2015 Posted July 22, 2015 the speaker has constant directivity across the entire audio band (good luck designing one) dipoles (or omni), or horns ... can get pretty well controlled directivity .... but sure, they're not your typical speaker. Your longer response, is spot on IMO.... the discussion of 'frequency response' in room, is quite complex. People often look to make a conclusion based on a single, or very limited selection, of charts, and this can be a mistake.
Pieface Posted July 24, 2015 Author Posted July 24, 2015 I should get to run some measurements suitable for windowing/gating this weekend and see what is happening with the "raw" speaker response from 250-300Hz up if I understand the process. I'm not sure @@svenr will be satisfied with my level of application to the cause but it's possible I may not be divorced at the conclusion of the measurements doing it my way I've been doing some additional reading on SBIR as I was still very punch drunk after my previous encounter with it. I think I was possibly using the front wall reflection formula for my sidewall/ceiling calculations which put me into a terminal spin trying to reconcile the measurement data to the calculated results. I seem to have a much closer correlation occuring between my calculations and the measured data now. Based on this I think I can have a play with moving my mains further from the wall and running a higher cross over point and putting the mains closer to the walls and damping the now higher SBIR affected frequencies and be able to make some kind of sense out of what I am seeing. [put's on devil's advocate hat] I note none of our gurus have managed to post any of their in room data anywhere that I can see. I really think actual results from someone who has a handle on this stuff and an explanation of what they have achieved, what compromises they made etc would be of a lot of value to mug punters exploring this information. It just feels a bit odd that there seems to be smoke and mirrors in terms of results of the applied theory around the objectivist approach to improving our in room loudspeaker performance in a typical home hifi situation. Imagine my horror if the reality is that unless you kit your listening space out like some sort of control room the results are as variable as a subjective approach of plonking speakers/furnishings etc where it "sounds good" or "looks good" :o 1
Recommended Posts