Isemcon EMX-7150 is an often recommended mic for higher SPL work.
https://www.isemcon.net/ashop/datasheets/EMX7150-US-r07-USletter.pdf
One of my better audio related purchases.
I guess it depends on the energy in the 1 sample. I can easily hear and measure the SPL of a single sample dirac pulse, but I doubt that the SPL measurement is close to right because of integration time being so much longer than the one pulse....iow, the SPL of the pulse gets averaged down. (if i'm thinking straight about REW's LZpeak...that it has a 35ms time weighting)
$320. Not cheap, but a fair amount less than some of the measurement mics that I've seen recommended here on DIY (Earthworks for example)
I think however much energy is in one sample 0.02ms @ 48kHz will be time averaged in over the SPL meter's integration period. (which can be pretty loud when it's a dirac pulse cranked up)
The SPL integration period is 35ms in Impulse mode in ARTA, and same for this $$$$ NTI SPL meter https://www.nti-audio.com/en/products/xl2-sound-level-meter/functions.
Impulse mode (35ms) is a shorter SPL weighting period than Slow (1s) or Fast (125ms).
So what i think Art and I were discussing is how would a peak of extremely short duration, (I used a 1 sample example) read on REW's peak meter.
Don't understand how accumulated loudness fits in the meter talk, unless you are talking about total accumulation throughout a meter's integration period.
The EAW SMAART V6.1 (I still use) has three "speeds", Slow, Fast, and Impulse, each of which has specific time constants.
REW has two "speeds", Slow and Fast, and an additional L peak.
If REWs "LZpeak" was "Impulse", it would not conform to the usual definition, and would also not read +3dB (the peak to average ratio)over "LZfast max" when measuring a continuous sine wave (peak to RMS ratio of 3dB) .
Last edited:
What is the usual definition of LZpeak? I can't find anything for it; LZ Impulse using 35 ms is all i've found anywhere for peaks.
I think REW's LZfast max and min will read sine wave RMS, so LZ peak should read +3dB...
So I'm still thinking REW uses a 35ms weighting for it's peak.
Ivan's NTI meter does, and it seemed to read a bit higher than our anecdotal hand clap comparisons using REW.
Like that means something, guess he claps louder. Lol.
Time to ask John M i guess.
I need a mulligan 
I'm seeing now that I misinterpreted the Data.
I'll try again
Before, I took the curve of the graph as presentation of SPL as well as crest factor. In reality the crest factor increases because generally speaking, RMS decreases as we increase HF, so of course Crest factor increases, in regards to music, Duh Duh Duh.
That doesn't explain why M noise has receding energy towards HF.... Then again true peak was 15db higher than the peak I can see on the RTA for M noise. True peak is 17db higher than the peak reading of the RTA regarding pink noise.
Missing the true peak versus frequency is once again, leaving no full context.
.
(M noise)
Pink noise holds a level/even peak across the spectrum as far as my RTA can see....
and that is more realistic than brown noise.....When using Pink noise to judge peak, RMS is irrelevant, thus, crest factor is irrelevant.....This whole thing is sorta stupid lol, in particular without knowing true levels vs frequency.
I have a true peak meter and a vst RTA
A high pass at 2.5khz showed the RTA highest peak to be 0-2db lower than the True peak meters highest reading. (true Peak max= -8db
A high pass at 1000hz showed true peak to be 3-4db above the highest Peak the RTA could capture. (true Peak max = -7db
A 120hz HP and 2.5khz LP showed the true peak to be 10-7db above the RTA's highest peak. (true Peak max=-4db
A low pass at 120hz LP showed true peak and the RTA's highest peak reading to be within 0-2db of each other.
Of course True peak reading highest. (true Peak max=-3db
RMS for the bass note comes within 5db of of true peak at times.....
This information is only so helpful as we do not have have the exact trend of where the true peaks land versus frequency...But sectioning of the spectrum into many parts might be a way of creating more detail. And then this is just One song....A bass heavy rap song.
Peak and true Peaks is within +/- ~3.5db..... Making Pink noise better than M noise for judging Headroom.
Peak is no less subject to the ideas of scientific sound quality than FR. A peak that is grossly higher than the other peaks will stand out very much.....Too much, if aiming for a balance, as music/movies tends to do. In particular, music that is meant for the louder listening crowd (EDM/RAP/ROCK).... If passage A is much quieter than passage B, in a classical piece. The peaks will still keep within a margin of error of each other.... The difference between Passage A and B will be similar to turning up or down, the gain knob.... Or the woodwinds will be lower than the brass section for Passage C..... so the peaks of the woodwinds will be in balance with each other, while the peaks of the brass section will be in balance with its selse while sitting on a higher shelf above the woodwinds. There will be more dynamic situations if one looks for them. A movie will on average be more dynamic, as whole....Think of a scene where the camera is viewing a very quiet field in the morning, and then a gun shot goes off as the hunter shoots at its target, the camera being his point of view..... It seems that in movies, no one wants to recreate the real dynamics of a chainsaw or gun shot btw....thats pretty important.
So thinking of the 115db Fork drop....Or 110db chain saw at 1meter...this establishes a basis for what real events can do....
We tend to monitor real life events at lower than real life level. Except in clubs lol. Peak is only as good as fast its capturing method.
So I recorded my clap below, shown in peak and 1/3rd oct
True peak is 12db higher than what my RTA could capture. -13db vs -1db.
Placing a high pass at 3khz, created a -15db peak on the RTA vs -2.6 on the True peak meter
In real life this clap was about 120db meaning I'd have to crank the stereo to achieve it.
Another error I made was regarding the Adjust RTA levels....One should leave this option off....for truest spl reading... consider that in The RTA things are more or less weighted.
https://www.roomeqwizard.com/help/help_en-GB/html/spectrum.html
In other words RTA and Spectrum is really slow compared to the Lpeak. Actually its so slow it is comparable to RMS even though its called Peak within REW RTA with the Spec
Transients in the lower frequency might might be easier to catch. The greatest difference between rta peak and true peak in the test above happened in the mid range during a certain kick that come in periodically. The kick seemed to peak near 100hz at -15db, causing the True peak meter to hit -4db at times.
This is what a loud listening session looks like on the 4" two ways at almost 1m, using the same song.
It was that damn kick that drove peak to 111db Zpeak....and it likely lives at 100hz.....
So after all of that analysis in 1/3octave, the hook line and sinker is that Z scale, is the actual most neutral, way to look at the system.
Below is pink noise on 1octave scaling...and please double check my work. Just know that in this setting, white noise appears flat.
the peaks of the music/movies are likely going to fall within a margin of error along this line, because everything is based off of neutral tone, with things typically falling with a range of each other. I made a prediction of those ranges above. I just noticed, the thickness of Meyer trend lines support my thoughts on my observations though I have only sampled a portion of what music could be. The thickest parts of the grey trends are about +/-3.5db....
You will also notice that in the sub area RMS and Peak can be very close....in Movies and Bass heavy music. Bass being, a lot of times, the most stressed system...Music/movies seek to maximize efforts here, regardless peak or RMS, conscience that there is only so much, yet desiring max fun factor....
I can place the idea of max headroom at the lowest note thanks to how the human ear works. The measurements we take, like a sine sweep, is akin to white noise. The same is true about the FR plots we simulate using HornResp. But everything we listen to, Music to Music and Games....etc etc, are all designed towards pink noise, or neutral sound to the ear, in so many words,
Peak expectations of 200hz can be generalized at 105db when the sub bass is hitting towards 115db when thinking about music or movies playing in this 1/1 scaling.
This is misleading
Think of the trend line now being the trend line of the RMS, in this situation Peaks will be the Pink line, maybe make the pink line a little more neutral.
These graphs are in 1/3rd octave so still deceiving us, in comparison to the view we normally judge our systems response by...1/1octave where white noise is flat and pink noise has a descending slope as we move to the right of the graph.
A 1/3rd Octave sine sweep is in order instead of only a 1/1 Option! I guess one could apply a filter in the shape of the pink noise slope on the master channel, before running the sweep.
The 115db fork drop recreation is going to ask of realistic levels for everything else if recorded properly. Meaning that not much else in the recording is going to top the 115db recording unless they happened to record something louder....Trying to listen to recordings of gun shots at realistic levels should record the footsteps at realistic levels unless the transfer function is messed up....For material gone through post production, done right, there will be an attempt at a 0 transfer function, as to sound natural as possible. We don't want to hear the microphone, in so many words. Music and movies definitely have this in mind during post production. Now its really a matter of how much higher, over the normal peak range, would we put an event. Rarely will you see any other peak, over Sub bass, while viewing in one octave FFT. Sometimes this is true of material viewed in 1/3ord octave
I pulled up "The Hulk" and checked out some loud areas
Notice the trend of peaks read by the RTA are as I described on this 1/3oct RTA. Plus minus 3.5db, in a round about way..... On the 1/1 FFT, this will be slanted like pink noise. The true peak meter maxed out at -4db which as about 6db than highest peak on the RTA......
One octave above vs 1/3 octave below, through the small two ways voiced with a slight bass tilt already.
Heres some white noise (black line) played under the movie curve, in 1/1 octave
The bass roll off of the 4" woofer is deceiving....Bass will continue to increase completing the trending line, on the 1/1 oct scale , using a system that is voiced neutral and more capable of covering the whole spectrum.
As can be seen, there will be much asked of the subwoofer by the time mid range ever reached 115db, more than most any home system could give....Using material that is post produced, as commonly seen
Subwoofer can be used for the headroom cap for most professionally produced media, mainly Music and Movie material. Video games will likely fall into the same fold.
here is the Action Movie "The Hulk" The vast amount of stimulus fills out the peak range evenly.
Here is todays chosen song....The areas where there are no main instrument fundamentals, lag behind in volume, ~450hz area and ~3500hz area
Switching the move or song into 1/1 octave moves all the peaks lower on the HF side.
Using these ideas I should be able to make better use of the system when deciding how much headroom is needed.
I'm seeing now that I misinterpreted the Data.
I'll try again
Before, I took the curve of the graph as presentation of SPL as well as crest factor. In reality the crest factor increases because generally speaking, RMS decreases as we increase HF, so of course Crest factor increases, in regards to music, Duh Duh Duh.
That doesn't explain why M noise has receding energy towards HF.... Then again true peak was 15db higher than the peak I can see on the RTA for M noise. True peak is 17db higher than the peak reading of the RTA regarding pink noise.
Missing the true peak versus frequency is once again, leaving no full context.
.
(M noise)
Pink noise holds a level/even peak across the spectrum as far as my RTA can see....
and that is more realistic than brown noise.....When using Pink noise to judge peak, RMS is irrelevant, thus, crest factor is irrelevant.....This whole thing is sorta stupid lol, in particular without knowing true levels vs frequency.
I have a true peak meter and a vst RTA
A high pass at 2.5khz showed the RTA highest peak to be 0-2db lower than the True peak meters highest reading. (true Peak max= -8db
A high pass at 1000hz showed true peak to be 3-4db above the highest Peak the RTA could capture. (true Peak max = -7db
A 120hz HP and 2.5khz LP showed the true peak to be 10-7db above the RTA's highest peak. (true Peak max=-4db
A low pass at 120hz LP showed true peak and the RTA's highest peak reading to be within 0-2db of each other.
Of course True peak reading highest. (true Peak max=-3db
RMS for the bass note comes within 5db of of true peak at times.....
This information is only so helpful as we do not have have the exact trend of where the true peaks land versus frequency...But sectioning of the spectrum into many parts might be a way of creating more detail. And then this is just One song....A bass heavy rap song.
Peak and true Peaks is within +/- ~3.5db..... Making Pink noise better than M noise for judging Headroom.
Peak is no less subject to the ideas of scientific sound quality than FR. A peak that is grossly higher than the other peaks will stand out very much.....Too much, if aiming for a balance, as music/movies tends to do. In particular, music that is meant for the louder listening crowd (EDM/RAP/ROCK).... If passage A is much quieter than passage B, in a classical piece. The peaks will still keep within a margin of error of each other.... The difference between Passage A and B will be similar to turning up or down, the gain knob.... Or the woodwinds will be lower than the brass section for Passage C..... so the peaks of the woodwinds will be in balance with each other, while the peaks of the brass section will be in balance with its selse while sitting on a higher shelf above the woodwinds. There will be more dynamic situations if one looks for them. A movie will on average be more dynamic, as whole....Think of a scene where the camera is viewing a very quiet field in the morning, and then a gun shot goes off as the hunter shoots at its target, the camera being his point of view..... It seems that in movies, no one wants to recreate the real dynamics of a chainsaw or gun shot btw....thats pretty important.
So thinking of the 115db Fork drop....Or 110db chain saw at 1meter...this establishes a basis for what real events can do....
We tend to monitor real life events at lower than real life level. Except in clubs lol. Peak is only as good as fast its capturing method.
So I recorded my clap below, shown in peak and 1/3rd oct
True peak is 12db higher than what my RTA could capture. -13db vs -1db.
Placing a high pass at 3khz, created a -15db peak on the RTA vs -2.6 on the True peak meter
In real life this clap was about 120db meaning I'd have to crank the stereo to achieve it.
Another error I made was regarding the Adjust RTA levels....One should leave this option off....for truest spl reading... consider that in The RTA things are more or less weighted.
https://www.roomeqwizard.com/help/help_en-GB/html/spectrum.html
In other words RTA and Spectrum is really slow compared to the Lpeak. Actually its so slow it is comparable to RMS even though its called Peak within REW RTA with the Spec
Transients in the lower frequency might might be easier to catch. The greatest difference between rta peak and true peak in the test above happened in the mid range during a certain kick that come in periodically. The kick seemed to peak near 100hz at -15db, causing the True peak meter to hit -4db at times.
This is what a loud listening session looks like on the 4" two ways at almost 1m, using the same song.
It was that damn kick that drove peak to 111db Zpeak....and it likely lives at 100hz.....
So after all of that analysis in 1/3octave, the hook line and sinker is that Z scale, is the actual most neutral, way to look at the system.
Below is pink noise on 1octave scaling...and please double check my work. Just know that in this setting, white noise appears flat.
the peaks of the music/movies are likely going to fall within a margin of error along this line, because everything is based off of neutral tone, with things typically falling with a range of each other. I made a prediction of those ranges above. I just noticed, the thickness of Meyer trend lines support my thoughts on my observations though I have only sampled a portion of what music could be. The thickest parts of the grey trends are about +/-3.5db....
You will also notice that in the sub area RMS and Peak can be very close....in Movies and Bass heavy music. Bass being, a lot of times, the most stressed system...Music/movies seek to maximize efforts here, regardless peak or RMS, conscience that there is only so much, yet desiring max fun factor....
I can place the idea of max headroom at the lowest note thanks to how the human ear works. The measurements we take, like a sine sweep, is akin to white noise. The same is true about the FR plots we simulate using HornResp. But everything we listen to, Music to Music and Games....etc etc, are all designed towards pink noise, or neutral sound to the ear, in so many words,
Peak expectations of 200hz can be generalized at 105db when the sub bass is hitting towards 115db when thinking about music or movies playing in this 1/1 scaling.
This is misleading
Think of the trend line now being the trend line of the RMS, in this situation Peaks will be the Pink line, maybe make the pink line a little more neutral.
These graphs are in 1/3rd octave so still deceiving us, in comparison to the view we normally judge our systems response by...1/1octave where white noise is flat and pink noise has a descending slope as we move to the right of the graph.
A 1/3rd Octave sine sweep is in order instead of only a 1/1 Option! I guess one could apply a filter in the shape of the pink noise slope on the master channel, before running the sweep.
The 115db fork drop recreation is going to ask of realistic levels for everything else if recorded properly. Meaning that not much else in the recording is going to top the 115db recording unless they happened to record something louder....Trying to listen to recordings of gun shots at realistic levels should record the footsteps at realistic levels unless the transfer function is messed up....For material gone through post production, done right, there will be an attempt at a 0 transfer function, as to sound natural as possible. We don't want to hear the microphone, in so many words. Music and movies definitely have this in mind during post production. Now its really a matter of how much higher, over the normal peak range, would we put an event. Rarely will you see any other peak, over Sub bass, while viewing in one octave FFT. Sometimes this is true of material viewed in 1/3ord octave
I pulled up "The Hulk" and checked out some loud areas
Notice the trend of peaks read by the RTA are as I described on this 1/3oct RTA. Plus minus 3.5db, in a round about way..... On the 1/1 FFT, this will be slanted like pink noise. The true peak meter maxed out at -4db which as about 6db than highest peak on the RTA......
One octave above vs 1/3 octave below, through the small two ways voiced with a slight bass tilt already.
Heres some white noise (black line) played under the movie curve, in 1/1 octave
The bass roll off of the 4" woofer is deceiving....Bass will continue to increase completing the trending line, on the 1/1 oct scale , using a system that is voiced neutral and more capable of covering the whole spectrum.
As can be seen, there will be much asked of the subwoofer by the time mid range ever reached 115db, more than most any home system could give....Using material that is post produced, as commonly seen
Subwoofer can be used for the headroom cap for most professionally produced media, mainly Music and Movie material. Video games will likely fall into the same fold.
here is the Action Movie "The Hulk" The vast amount of stimulus fills out the peak range evenly.
Here is todays chosen song....The areas where there are no main instrument fundamentals, lag behind in volume, ~450hz area and ~3500hz area
Switching the move or song into 1/1 octave moves all the peaks lower on the HF side.
Using these ideas I should be able to make better use of the system when deciding how much headroom is needed.
Attachments
Last edited:
I placed the cursor at the highest 2nd order thd before -15db of ~200hz.....5.5% at 215hz...
The woofer is dominate according to VituixCad. Where the Horn , is down 15, the mid woofer carries the signal.
After attenuating the 2 peaks in the Horns roll off, in addition to a lr48 at 183hz, The driver is even lower at 200hz, about -17db, and everything below is receives a sharp reduction with introducing a higher GD to the system. The mid woofer has a LR24 low pass at 300hz. A KA of 1.5 points to 429hz which also happens to be -15db down on the midwoofer, where the horn is down 2-3db and dominating. So That means that I can be on axis with the horn without worrying about getting out of the sweet spot of the mid woofer. Even at 1 meter. I'd need to take a measurement with the mic in horn resting on the exit, at volume, to see what thd says below horn resonance in order to avoid excursion strain that will contaminate output of everything else.
These diagonal lines are approximate pink noise slopes on the 1/1 scaled graph, plenty of headroom, with 350hz achieving 1.4% 2nd at 108db at one meter. 215hz had the highest 2nd order thd above -15db on the first measurement. 5.5% at 97,7db..... This happens to remain 97db with the XO
If you remember the Rap song that was balanced within 3.5db plus/minus on the 1/3oct...here it is on the a 1/1 scale
vs
Attachments
Last edited:
If you want to make high SPL measurements cheaply you can use a dynamic microphone, generating a cal file using a known good measurement microphone and a speaker. Won't be the last word in accuracy but good enough for distortion measurements, compression measurements etc. (for compresion no need to even make a cal file).
Very fair. There’s a translated copy of a review with tons of measurements of the microphone’s limits and polar pattern here:
https://www.isemcon.com/datasheets/iSEMcon EMX-7150-US.pdf
It has a very low sensitivity, so care must be taken to match the audio interface or preamp as well. Outside of distorting the mic capsule - most cheap electret condenser measurement mics publish an SPL max of ~130dB with 3% THD at 1kHz pure tone, so with broadband signals like pink noise at 12dB CF you have perhaps 115dB capability before the mic starts to contaminate your data - it’s very common to be driving the preamp to distortion, or be sat in the noise floor for low level measurements by comparison.
http://www.sengpielaudio.com/calculator-transferfactor.htm
It’s surprising how few ‘audio engineers’ consider the basics of gain structure/sensitivity/impedance in their signal chain, even for recording.
Also @camplo I appreciate you’re trying to understand but you’re yet again sort of reinventing the wheel. If you read the M-Noise or AES75-2022 test procedure document, you’ll see that it’s quite easy to find out whether your system has suitable headroom for music reproduction.
https://www.aes.org/standards/blog/2022/3/aes75-2022-published
You do however need a relatively large space, FFT based software with certain conditions in processing, a heck of a lot of unclipped peak voltage output at the amplifier and a mic/preamp combination that can handle the expected peak levels.
Your UMIK is not suitable, unfortunately. Heck, even my isemCON mics and a bridged Powersoft X8 channel pair at 400V peak output aren’t suitable to complete the M-Noise testing procedure for some of the PA kit I use.
@weltersys
Art, I'm replying to my own post in order to correct it, as I've confirmed that you were right....John M kindly replied to my inquiry on the REW forum.
His reply, "The peak reading is based on the largest absolute sample value seen in each block of audio data. There is no time weighting."
That's really cool and makes the meter even more valuable imo.
I forgot to write some stuff about the Fulcrum/Gunness approach to FIR for temporal ‘fixes’ earlier.
Without going into too much detail, they aren’t using simple IR inversions from an acoustic measurement. That simple inversion process is easy to do, and it makes a pretty graph but rarely a good sounding speaker. Unless you stay fixed in a very small physical position, never turn it up (or down), the parts don’t age
An electrical filter is a 1D solution, so care has to be taken to determine the root cause of any measured or heard phenomena, and then that needs to be isolated. In short, if the individual ‘problem’ is not purely two-port systems, then filters are only going to be effective in one position - and often make things worse everywhere else.
“tq” or “Focusing” is a multi-stage process, with several different filters being convolved for the final result. I’ve installed a fair few Fulcrum boxes now, and they all got opened up to look inside. Measurements of the cabinet with and without processing, plus measurement of the amplifier outputs helped switch on a few lightbulbs.
The original Gunness patent clarifies the actual steps in more detail, and other folks’ papers expand on it further. I’ve had chats with various folk in the pro audio world about this over the years, and the process typically involves measuring a lot of devices and making/refining lumped models of the behaviour to ensure it’s ‘fixable’. The Wavelet Transform is a very useful tool for this... Each filter is then typically synthesised to specified parameters, and combined with other ‘dark art’ to suit the manufacturer’s needs.
If you’re not able to script this yourself or use existing tools like the LEM section of Akabak to model things like compression driver phase plug reflections, then I can only recommend averaging a lot of anechoic measurements at various angles, drive levels, and over multiple driver samples. The choice of averaging method is up to you depending on what you want to fix. Averager from Eclipse Audio is free and has all the options you'd generally need for that:
https://eclipseaudio.com/averager/
As to “why?” the ‘flat’ mag & phase response makes integrating multiple speakers of different size, type and location a lot easier. It does also sound ‘better’ than the pure IIR equivalent on the same device at higher drive levels in my experience, although I’ve not yet been able to set up a ‘proper’ double blind test.
I might have shared these before, but for the readers among us:
Greenfield, R. R. and Hawksford, M. J. (1989) “Efficient filter design for loudspeaker equalization,” in 86th AES Convention.
Norcross, S. G., Soulodre, G. A. and Lavoie, M. C. (2004) “Subjective investigations of inverse filtering,” Journal of the Audio Engineering Society. Audio Engineering Society, 52(10), pp. 1003–1028.
Gunness, D. W. and Hoy, W. R. (2005) “A spectrogram display for loudspeaker transient response,” in Audio Engineering Society Convention 119. Audio Engineering Society. Available at: https://www.aes.org/e-lib/browse.cfm?elib=13339
And
https://patents.google.com/patent/US8081766B2/en?inventor=david+gunness&oq=david+gunness
Ponteggia, D. and Di Cola, M. (2007) “Time-Frequency Characterization of Loudspeaker Responses Using Wavelet Analysis,” in Audio Engineering Society Convention 123. Audio Engineering Society. Available at: https://www.aes.org/e-lib/browse.cfm?conv=123&papernum=7204.
Di Cola, M. and Ponteggia, D. (2010) “Application of optimized Inverse Filtering to improve time response and phase linearization in multi-way loudspeaker systems,” in Audio Engineering Society Convention 129. Audio Engineering Society. Available at: http://www.aes.org/e-lib/download.cfm?ID=15683
Vaucher, R. (2013) “Linear phase implementation in loudspeaker systems: Measurements, processing methods, and application benefits,” in Audio Engineering Society Convention 135. Audio Engineering Society. Available at: http://www.aes.org/e-lib/download.cfm?ID=16976
All of this is getting very much ahead of where camplo's current setup is, though! I'd strongly suggest getting the minimum phase stuff right first with simple IIR filters, before you consider messing with FIR.
Without going into too much detail, they aren’t using simple IR inversions from an acoustic measurement. That simple inversion process is easy to do, and it makes a pretty graph but rarely a good sounding speaker. Unless you stay fixed in a very small physical position, never turn it up (or down), the parts don’t age
An electrical filter is a 1D solution, so care has to be taken to determine the root cause of any measured or heard phenomena, and then that needs to be isolated. In short, if the individual ‘problem’ is not purely two-port systems, then filters are only going to be effective in one position - and often make things worse everywhere else.
“tq” or “Focusing” is a multi-stage process, with several different filters being convolved for the final result. I’ve installed a fair few Fulcrum boxes now, and they all got opened up to look inside. Measurements of the cabinet with and without processing, plus measurement of the amplifier outputs helped switch on a few lightbulbs.
The original Gunness patent clarifies the actual steps in more detail, and other folks’ papers expand on it further. I’ve had chats with various folk in the pro audio world about this over the years, and the process typically involves measuring a lot of devices and making/refining lumped models of the behaviour to ensure it’s ‘fixable’. The Wavelet Transform is a very useful tool for this... Each filter is then typically synthesised to specified parameters, and combined with other ‘dark art’ to suit the manufacturer’s needs.
If you’re not able to script this yourself or use existing tools like the LEM section of Akabak to model things like compression driver phase plug reflections, then I can only recommend averaging a lot of anechoic measurements at various angles, drive levels, and over multiple driver samples. The choice of averaging method is up to you depending on what you want to fix. Averager from Eclipse Audio is free and has all the options you'd generally need for that:
https://eclipseaudio.com/averager/
As to “why?” the ‘flat’ mag & phase response makes integrating multiple speakers of different size, type and location a lot easier. It does also sound ‘better’ than the pure IIR equivalent on the same device at higher drive levels in my experience, although I’ve not yet been able to set up a ‘proper’ double blind test.
I might have shared these before, but for the readers among us:
Greenfield, R. R. and Hawksford, M. J. (1989) “Efficient filter design for loudspeaker equalization,” in 86th AES Convention.
Norcross, S. G., Soulodre, G. A. and Lavoie, M. C. (2004) “Subjective investigations of inverse filtering,” Journal of the Audio Engineering Society. Audio Engineering Society, 52(10), pp. 1003–1028.
Gunness, D. W. and Hoy, W. R. (2005) “A spectrogram display for loudspeaker transient response,” in Audio Engineering Society Convention 119. Audio Engineering Society. Available at: https://www.aes.org/e-lib/browse.cfm?elib=13339
And
https://patents.google.com/patent/US8081766B2/en?inventor=david+gunness&oq=david+gunness
Ponteggia, D. and Di Cola, M. (2007) “Time-Frequency Characterization of Loudspeaker Responses Using Wavelet Analysis,” in Audio Engineering Society Convention 123. Audio Engineering Society. Available at: https://www.aes.org/e-lib/browse.cfm?conv=123&papernum=7204.
Di Cola, M. and Ponteggia, D. (2010) “Application of optimized Inverse Filtering to improve time response and phase linearization in multi-way loudspeaker systems,” in Audio Engineering Society Convention 129. Audio Engineering Society. Available at: http://www.aes.org/e-lib/download.cfm?ID=15683
Vaucher, R. (2013) “Linear phase implementation in loudspeaker systems: Measurements, processing methods, and application benefits,” in Audio Engineering Society Convention 135. Audio Engineering Society. Available at: http://www.aes.org/e-lib/download.cfm?ID=16976
All of this is getting very much ahead of where camplo's current setup is, though! I'd strongly suggest getting the minimum phase stuff right first with simple IIR filters, before you consider messing with FIR.
Last edited:
To me there is a different side to this.
Way back when, I was more interested in making waveguides that did not have the later time artifacts than Gunness shows. Hence, there would be no need for the complex filters which make up his patent. Given knowledge back than, I suppose his approach was enlightening, but these days, there should not be anything in the devices field responses that need that kind of correction.
Way back when, I was more interested in making waveguides that did not have the later time artifacts than Gunness shows. Hence, there would be no need for the complex filters which make up his patent. Given knowledge back than, I suppose his approach was enlightening, but these days, there should not be anything in the devices field responses that need that kind of correction.
The 35ms time weighting of Impulse response seems downright slow when you look at something like finger snaps- the entire envelope before reflections is only 6ms.@weltersys
Art, I'm replying to my own post in order to correct it, as I've confirmed that you were right....John M kindly replied to my inquiry on the REW forum.
His reply, "The peak reading is based on the largest absolute sample value seen in each block of audio data. There is no time weighting."
That's really cool and makes the meter even more valuable imo.
And finger snaps are really slow when compared to compression driver phase plug reflections, as Kyleneuron just mentioned in regards to the "Gunness Focusing" approach to FIR temporal correction.
I completely agree about first getting the minimum phase stuff right with IIR filters, before messing with FIR- learn to ride a bike before attempting to drive a rocket powered unicycle
Art