Hi All
The other day I came to think of Bob Carver and his ”Transfer funktion duplication” as described in this Wikipedia article:
https://en.wikipedia.org/wiki/Bob_Carver se ”Amplifier Modeling”.
I wonder, if something similar could be transferred to Loudspeakers with DSP? I plan to build a pair of really big MEH´s for corner-placement, and will use DSP to manage crossover and EQ. But what target-curve/sound/coloring should I choose? That´s what I want this thread to be about. “Soundshaping”!
I think @wesayso has done some of what I am asking about with his line-array-speakers, duplicating some Altec-sound as heard at his neighbor, if I remember correctly.
So, what do you do and how? What makes a loudspeaker sound as it does? I know this is not simple, but much can be done with DSP!? And of course some things need to be done in the hardware like directivity, placement in the room …....
If there is an existing thread on this topic please let me know.
This is my first thread on diyaudio after many years of “recreational reading” and gathering of knowledge and ideas. I don´t have much to add my self as of now.
In curious expectation
Steffen
The other day I came to think of Bob Carver and his ”Transfer funktion duplication” as described in this Wikipedia article:
https://en.wikipedia.org/wiki/Bob_Carver se ”Amplifier Modeling”.
I wonder, if something similar could be transferred to Loudspeakers with DSP? I plan to build a pair of really big MEH´s for corner-placement, and will use DSP to manage crossover and EQ. But what target-curve/sound/coloring should I choose? That´s what I want this thread to be about. “Soundshaping”!
I think @wesayso has done some of what I am asking about with his line-array-speakers, duplicating some Altec-sound as heard at his neighbor, if I remember correctly.
So, what do you do and how? What makes a loudspeaker sound as it does? I know this is not simple, but much can be done with DSP!? And of course some things need to be done in the hardware like directivity, placement in the room …....
If there is an existing thread on this topic please let me know.
This is my first thread on diyaudio after many years of “recreational reading” and gathering of knowledge and ideas. I don´t have much to add my self as of now.
In curious expectation
Steffen
Let me first explain what I had to do to get decent sound in my room. With a line array you get a totally different effect in-room due to the drivers interaction with each other. Al low frequencies all drivers sum, and the higher one goes in frequency, the more the drivers interfere with each other, because the array becomes larger that the wave length they are playing. There is a whole bunch of math behind it that is explained in this thread. It results in a frequency graph that, by itself, is useless to sound good. The curve drops down by about 3 dB/octave towards the high frequencies.
An example of the output of an array (simulated data)
So to get something that sounds good, I had to use EQ. Having no idea how to go about it, I started reading a lot about "House curves".
I came across several interesting pieces, among which a curve as suggested by Bob Katz, for his studio work. he suggested a knee at a
certain frequency, from where the higher frequencies must be rolled off a bit. After even more research, I noticed that trend being present
in various house curves, only where the knee started could vary slightly. I started listening to a lot of different curves:
Curves as measured at the listening spot. That's where I measured and set my house curve.
After a few exercises like the above , I picked the one I liked best with lots of different types of music, and lived with it
for a while. Use it in every day use for a long period, while making little teaks every now and then. Slowly growing into
a curve that worked for most material. Having done that for months, I grew towards a certain curve that did good on
almost every song played. I only needed to keep the average SPL level the same too, as that influences the perception.
I chose a fixed average SPL level of about 85 to 87 dB on average. (meaning peaks are way higher)
I use JRiver to get my songs play at an equal average SPL figure. It uses an industry standard R128 algorithm to achieve
that, after first scanning each song (using the function: Analyze Audio) for it's dynamic range to set the proper SPL level.
After looking at the curve I grew into, I started comparing it to the various graphs of curves I had collected. One curve
stood out for me, as it was a near spitting image of what I had arrived on:
The dashed "trained listeners only" curve in this graph by Toole was as near as I could find to my preferred curve.
But, as mentioned by Steffen, I visited my next door neighbor with his Altec Barcelona speakers. He played some music
I knew well and I fell in love with (specifically the bottom half on vocals of) that sound. There was a touch of warmth to
it that just seemed to embrace you. But without muddying up the rest of the music. I started tweaking my curve and
ended up changing the 200 to 300 Hz range to get that touch of warmth. Add too much and everything sounds muddy.
Add just enough and you get that warm cozy sound I fell in love with...
That little tweak was gold for my listening pleasure, even though technically it probably isn't the most "transparant".
These days I write that off as being "My-Fi" instead of true "Hi-Fi". Anyway, that's my background of how I got there.
I do use other tweaks that I've written about, but for tonal balance I use the above graph with some small tweaks to
get where I want to be just that little bit faster.
If the above story was in-line with what you expected from starting this thread, may I suggest you read Mitch's review
of the "Dynaudio Focus 600"? Not because of those speakers, but about half way down there's a part that says:
"Objective Measurements - Introduction" where Mitch added several interesting links to papers that might prove
an interesting read.
An example of the output of an array (simulated data)
So to get something that sounds good, I had to use EQ. Having no idea how to go about it, I started reading a lot about "House curves".
I came across several interesting pieces, among which a curve as suggested by Bob Katz, for his studio work. he suggested a knee at a
certain frequency, from where the higher frequencies must be rolled off a bit. After even more research, I noticed that trend being present
in various house curves, only where the knee started could vary slightly. I started listening to a lot of different curves:
Curves as measured at the listening spot. That's where I measured and set my house curve.
After a few exercises like the above , I picked the one I liked best with lots of different types of music, and lived with it
for a while. Use it in every day use for a long period, while making little teaks every now and then. Slowly growing into
a curve that worked for most material. Having done that for months, I grew towards a certain curve that did good on
almost every song played. I only needed to keep the average SPL level the same too, as that influences the perception.
I chose a fixed average SPL level of about 85 to 87 dB on average. (meaning peaks are way higher)
I use JRiver to get my songs play at an equal average SPL figure. It uses an industry standard R128 algorithm to achieve
that, after first scanning each song (using the function: Analyze Audio) for it's dynamic range to set the proper SPL level.
After looking at the curve I grew into, I started comparing it to the various graphs of curves I had collected. One curve
stood out for me, as it was a near spitting image of what I had arrived on:
The dashed "trained listeners only" curve in this graph by Toole was as near as I could find to my preferred curve.
But, as mentioned by Steffen, I visited my next door neighbor with his Altec Barcelona speakers. He played some music
I knew well and I fell in love with (specifically the bottom half on vocals of) that sound. There was a touch of warmth to
it that just seemed to embrace you. But without muddying up the rest of the music. I started tweaking my curve and
ended up changing the 200 to 300 Hz range to get that touch of warmth. Add too much and everything sounds muddy.
Add just enough and you get that warm cozy sound I fell in love with...
That little tweak was gold for my listening pleasure, even though technically it probably isn't the most "transparant".
These days I write that off as being "My-Fi" instead of true "Hi-Fi". Anyway, that's my background of how I got there.
I do use other tweaks that I've written about, but for tonal balance I use the above graph with some small tweaks to
get where I want to be just that little bit faster.
If the above story was in-line with what you expected from starting this thread, may I suggest you read Mitch's review
of the "Dynaudio Focus 600"? Not because of those speakers, but about half way down there's a part that says:
"Objective Measurements - Introduction" where Mitch added several interesting links to papers that might prove
an interesting read.
Last edited:
You touched upon it in your original post. The directivity of the speakers influence what sounds good. Meaning that
the speaker directivity can determine if and how much the top end should drop down. So take the above as a starting
point and do further tweaks from there on. The basic shape can guide you, the exact values might differ in preference.
just look at how wide all the different preferences are within that graph from Toole. As you listen more and more, it
becomes easier to be more critical of what it is you're hearing. Live with it for longer periods, it is much harder to
determine what's right in a short brief moment. Yet, coming in with fresh ears the following day often is an eye opener
too!
A graph that helped me make sense of what I heard is the Interactive Frequency Chart. Hover the mouse over the
frequencies to see additional information of each EQ band. It helped me to learn to listen and know what I heard.
A static sample of what the chart looks like:
the speaker directivity can determine if and how much the top end should drop down. So take the above as a starting
point and do further tweaks from there on. The basic shape can guide you, the exact values might differ in preference.
just look at how wide all the different preferences are within that graph from Toole. As you listen more and more, it
becomes easier to be more critical of what it is you're hearing. Live with it for longer periods, it is much harder to
determine what's right in a short brief moment. Yet, coming in with fresh ears the following day often is an eye opener
too!
A graph that helped me make sense of what I heard is the Interactive Frequency Chart. Hover the mouse over the
frequencies to see additional information of each EQ band. It helped me to learn to listen and know what I heard.
A static sample of what the chart looks like:
Hi Steffen,
Yep, it appears Bob Carver was simply matching amplifiers' transfer functions. So he made the transfer functions of the $$$$ amps he wanted to replicate, the targets of the $ amps he built to match up.
The same can be done with speakers. Just make speaker A's transfer function the target of speaker B.
For example, FirDesigner has a tutorial for doing this....https://eclipseaudio.com/loudspeaker-matching/
The GIANT caveat is that unlike matching amplifier which have no spatial signal properties, speakers need to have very similar radiation patterns for the technique to work very well.
And there's also the big issue of which transfer function do you match to ?....on-axis, some polar average, etc ?
But given fairly similar boxes, it can work decently well. A Music Tribe X/M-32 mixer using their particular speakers which designed to be connected via Cat5 on Ultranet (a private type ethernet), has presets for various speaker models to emulate other manufactures speakers that are widely used in live sound.
I have a Music Tribe IQ-15 speaker, that has a x-32 preset to emulate a Meyer UPA-1p which i also have. Emulation worked a little better than expected.
I'v never tried the FirD routine though, as i don't really care enough to bother...(i kinda like that things sound different !)
One thing i can offer for sure...if you make large syns, they will act quite differently in the room corners vs elsewhere. Corners shift the tonal balance lower pretty substantially. I like to first tune my snys to flat where ever i put them, and then start playing with house curves by ear. I've found the idea of a universal house curve, like the typical on-axis downward-sloper, to be a step in the right direction, but doesn't take room placement and radiation pattersn into account well enough.
Yep, it appears Bob Carver was simply matching amplifiers' transfer functions. So he made the transfer functions of the $$$$ amps he wanted to replicate, the targets of the $ amps he built to match up.
The same can be done with speakers. Just make speaker A's transfer function the target of speaker B.
For example, FirDesigner has a tutorial for doing this....https://eclipseaudio.com/loudspeaker-matching/
The GIANT caveat is that unlike matching amplifier which have no spatial signal properties, speakers need to have very similar radiation patterns for the technique to work very well.
And there's also the big issue of which transfer function do you match to ?....on-axis, some polar average, etc ?
But given fairly similar boxes, it can work decently well. A Music Tribe X/M-32 mixer using their particular speakers which designed to be connected via Cat5 on Ultranet (a private type ethernet), has presets for various speaker models to emulate other manufactures speakers that are widely used in live sound.
I have a Music Tribe IQ-15 speaker, that has a x-32 preset to emulate a Meyer UPA-1p which i also have. Emulation worked a little better than expected.
I'v never tried the FirD routine though, as i don't really care enough to bother...(i kinda like that things sound different !)
One thing i can offer for sure...if you make large syns, they will act quite differently in the room corners vs elsewhere. Corners shift the tonal balance lower pretty substantially. I like to first tune my snys to flat where ever i put them, and then start playing with house curves by ear. I've found the idea of a universal house curve, like the typical on-axis downward-sloper, to be a step in the right direction, but doesn't take room placement and radiation pattersn into account well enough.
Last edited:
Idk to discuss Bobs dubious marketing as scientific principle and then apply it to speakers is apocryphal. Fine as a thought experiment, but bobs conjecture isn’t even a universally accepted theory/principle for amplifiers (who’s performance has a higher universal standard). No amount of filters, dsp or ‘bass eq’ will change that. In addition imo too much processing decreases musicality and increases fatigue.
A little, and it will depend on the DSP.
It's perfectly possible to match the on-axis frequency response of any speaker to any speaker, so long as neither speaker is hitting any limitations. For example, a 1" dome tweeter can have the same frequency response as an 18" subwoofer, but the 1" dome (with 0.1mm Xmax) will have very limited SPL available.
Things you can match with a more DSP than is typically used:
- The phase response. You'll need FIR filtering, but it is possible.
- Harmonic distortion. Here, it gets more tricky. If you start with a very clean speaker, even just trying to match the distortion profile of a not-so-clean speaker will be very difficult. You'll need some creative tools, like frequency multipliers (to create 2nd, 3rd, 4th etc harmonics), probably some compression/expansion, and before/after frequency response shaping. I'm pretty sure it could be done, but it would take a good sound engineer many hours to get anything close.
There's also the consideration that your speakers aren't perfectly clean, so we must find some way of accounting for that.
The end result would be interesting. I'm not sure anyone has really tried to that extent, and there's a whole market segment of people trying to replace guitar speakers with software.
Almost impossible:
- Directivity matching.
Here, life gets difficult. Really really difficult. Bang and Olufsen have got the closest, with their Beolab 90. They've done narrow/wide/360 directivity with those speakers, and haven't said what frequency range it works for. If you want to match the directivity response of a different speaker, you'll need measurements of the other speaker, and then a way of telling something-like-the-Beolab-90 about the target directivity pattern.
Chris
Look here: https://www.tonmeister.ca/wordpress/2015/11/04/bo-tech-intuitive-directivity-plots/ which is where some of the Beolab 90 plots can be found,
and in other articles written by Geoff Martin on his Ear fluff and Eye candy page.
When going for tonal balance tweaks, make sure you check out what your room does. Lots of early reflections will mess with the sound and will bury the tweaks quite fast (might even bring on that lack of musicality and listening fatigue). Don't blame the tools though. You can get great results with DSP, but you have to make sure you're starting with something good. A MEH is a very good start 👍. It will work with the room if it is placed well.
Hi and thanks for all the answers 🙂
It is really a big topic and with lots of different opinions on what is right. I like your term MyFi. What you wrote in your first post makes good sense to me, especially that it takes time and that it is highly subjective.
I have especially one old HiFi friend who keeps on stating that I NEVER EVER can build ANYTHING that sounds as good as what the professionals/the industri can build! Hmm, he keeps on talking about Sonus Faber loudspeakers, and that they design and produce such fantastic loudspeakers! And I am sure they do. But I guess I should remember to build for ME and not for HIM!
I am into DIY because I grew tired of all the pseudo-sciense and woodoo in the industry, all the gold-plated Bling Bling. I can´t really afford a ”High-End” loudspeaker either. Also I have learned enough to know that sound is a systems-approach, meaning that i need to integrate the speaker in the room with some kid of "room-correction/adjustments" anyway. Can´t just buy a super speaker and expect to have super sound!
On top of that, I can not buy what I really want to have: A really big MEH (like a SH96), a powerful point-source, as I have been listening to fullrange-speakers for many years. The corner-placement makes sense in my room I think, and saves space too. Some years ago I read a post by Tom Danley, that he could make his side-walls in a rather narrow room kind of disappear with his SH 50´s up against the sidewalls! That made me really curious, I am exited to find out.
Steffen
It is really a big topic and with lots of different opinions on what is right. I like your term MyFi. What you wrote in your first post makes good sense to me, especially that it takes time and that it is highly subjective.
I have especially one old HiFi friend who keeps on stating that I NEVER EVER can build ANYTHING that sounds as good as what the professionals/the industri can build! Hmm, he keeps on talking about Sonus Faber loudspeakers, and that they design and produce such fantastic loudspeakers! And I am sure they do. But I guess I should remember to build for ME and not for HIM!
I am into DIY because I grew tired of all the pseudo-sciense and woodoo in the industry, all the gold-plated Bling Bling. I can´t really afford a ”High-End” loudspeaker either. Also I have learned enough to know that sound is a systems-approach, meaning that i need to integrate the speaker in the room with some kid of "room-correction/adjustments" anyway. Can´t just buy a super speaker and expect to have super sound!
On top of that, I can not buy what I really want to have: A really big MEH (like a SH96), a powerful point-source, as I have been listening to fullrange-speakers for many years. The corner-placement makes sense in my room I think, and saves space too. Some years ago I read a post by Tom Danley, that he could make his side-walls in a rather narrow room kind of disappear with his SH 50´s up against the sidewalls! That made me really curious, I am exited to find out.
Steffen
Yes, I also know someone who says these things.
In a domestic situation you can have a single waveguide that goes below 1kHz using a single driver.
Last edited:
One notable attempt is Crutchfield's "SpeakerCompare" - where you upload your own music and via DSP magic, you can listen to how it sounds across various speaker models they sell, using headphones. Of course, the headphones used matters...
You can get a VST plug in for about $25, that does controllable levels of 2nd and 3rd harmonic distortion. Unsure how that would work on an entire finished recording, but for individual instrument tracks is allegedly does work well.
Finally, someone cooked up an AI based routine that can listen to how something modifies the sound of its audio input and generate a VST plug in that emulates that sound modification. Perhaps at this stage the resolution isnt up to what Bob had done with amplifier sonics, but me thinks that level of success is only a matter of time.
You can get a VST plug in for about $25, that does controllable levels of 2nd and 3rd harmonic distortion. Unsure how that would work on an entire finished recording, but for individual instrument tracks is allegedly does work well.
Finally, someone cooked up an AI based routine that can listen to how something modifies the sound of its audio input and generate a VST plug in that emulates that sound modification. Perhaps at this stage the resolution isnt up to what Bob had done with amplifier sonics, but me thinks that level of success is only a matter of time.
Just a quick note that I spent about 18 months working, full time, for a company that sells high-end HiFi speakers. £140k/pair was the top model.
I was building/testing/repairing those speakers, so I got to know them well.
Let me say this, then: A competent DIYer could make something better with £1000 spent on drivers.
It's also worth noting that some HiFi companies get it right, or at least do quite well. The one I worked for did not.
Chris
I like that ability to audition a speaker remotely like Crutchfield provides. Similar to Crutchfield this youTube speaker reviewer built a filter to represent the speaker on axis response and played music selections through it.
I
Imagine a "service" where;
Who'd do it?
- you download software that runs a test on your system (say just one channel and it assumes symmetry) with a test microphone / PC.
- you upload the results.
- anyone can then upload their own music track and hear what your system in the room sounds like, through headphones.
- the headphones used for playback matter. There's selectable filters for the recommended models also.
Who'd do it?
Hi Allan 🙂
I think I have to re-state, that my main objektive with this thread is to discuss, HOW to "manipulate" the sound/transfer-function of a capable Loudspeaker to a certain desired sound, maybe inspired by existing products.
It is definitely not so, that I want to copy a specific speaker (Like Bob copied the sound of other amplifiers to his own amplifier), then I could just buy that speaker.
It´s more about how subtle are the differences of the transfer-functions, How do they look on a frequency-chart, bandwidth and gain. And then maybe have options to pick and choose. Do I want warm sound, transparent/analytical, bright sound what ever characteristics.......
It´s really a new and exciting world of opportunities what you can do with software, DSP and plug-in´s!
Steffen
Bugger, nearly derailed my own thread!😳 It wasn´t my intend to discuss physical hardware or my specific project.
I think I have to re-state, that my main objektive with this thread is to discuss, HOW to "manipulate" the sound/transfer-function of a capable Loudspeaker to a certain desired sound, maybe inspired by existing products.
It is definitely not so, that I want to copy a specific speaker (Like Bob copied the sound of other amplifiers to his own amplifier), then I could just buy that speaker.
It´s more about how subtle are the differences of the transfer-functions, How do they look on a frequency-chart, bandwidth and gain. And then maybe have options to pick and choose. Do I want warm sound, transparent/analytical, bright sound what ever characteristics.......
It´s really a new and exciting world of opportunities what you can do with software, DSP and plug-in´s!
Steffen
I've read it's happened to people and I can claim I did it to someone at least once. Goes something like "This is supposed to sound like the speakers playing" ... "Ok, I'm ready (puts headphones on) turn the speakers off" ... "They are off" ... "What!?!"
While the 411A was a wonderful performing wide range woofer, it's at ~160 deg (theoretical 180 deg) out to ~800 Hz/15", so the 'warmth' is as much or more due to the cab than the driver based on comparison testing at the Altec S.E. distributor with the Barcelona Vs its 9846 studio monitor and my own comparisons with other various size speakers before them when folks started wanting me to downsize my typical pioneer's big/huge cabs to fit in dorm rooms and new, smaller apartments, yet keep their performance. Sound familiar? 🙂
In short, IME, if not wall/corner loaded, cabs need to be large enough to cover down to the driver's LF mass corner (Fhm = ~2*Fs/Qts') as part of the requirements to get a ~ 'feel'/'warmth'/'fullness' of a live performance within the obvious room, reproduction limitations.
Auralization, there is (expensive) software that can do it, Odeon, CATT-Acoustic, perhaps some others. One could do reasonably well with just Javascript, Web Audio Api, there are some auralization demos in github. Wouldn't be too difficult to make some generalized rooms, include some kind of HRTF stuff, and drop it all behind a URL anyone can test. Problem with this that stopped me doing it, was how to make realistic source out of set of loudspeakers. I have no idea how to implement the speaker output, to various directions with Web Audio API or in any software (I'm not pro on audio processing, can do Javascript though). Found some code for Web Audio API that has either point sources or some "ideal" patterns like cardioid though. To summarize, auralization is not too difficult for anyone to do with little bit of search / copy paste code. Problem seems to be making realistic sound sources out of loudspeaker (measurements). Erin for example has the data, speaker output to all directions, which I think he might be able to export from klippel as Balloon data that the Odeon etc. software eats. Data is there I think, we'd just need someone to convert that into virtual realm 🙂 Anyway, interesting stuff. Not sure if that would sound anywhere realistic but still might be somewhat good way to evaluate between speakers. At least it would be very very cool way to study some of the phenomena like early reflections, effect of coverage pattern and what not. Very very cool. Programmable room and speakers.
Steffen,
There is a way that may bring you in the direction you aim for, but there is a number of caveats: the drivers must be the same size as the original (off axis will then be roughly similar) plus the enclosure must have roughly the same dimensions (baffle effects having identical impact) : Copy the acoustic transfer function of each driver precisely and you will be close.
I predict many here will object strongly ("each driver has its own sound signature.."). Of course the drivers must not have strong break-up in their respective passbands.
There is an experiment by Member Jon Bocani with compression drivers: he concluded that, once minutely equalized, they sounded identical.
He got flamed here, but there was little hard evidence against his case. Furthermore, member MarkBakk posted a German experiment with equalized tweeters, that came to the same conclusion. Earl Geddes is also of the opinion compression drivers, regardless of price, are basically a commodity if they meet certain mininum specs: it is the waveguide that makes most of the difference.
One may have some questions, but it is an experiment I would like to undertake myself. Listening to pass band filtered midranges having the same size, housing, level and acoustic transfer function would be an interesting starting point.
Oops, now I have put my head in the guillotine..
There is a way that may bring you in the direction you aim for, but there is a number of caveats: the drivers must be the same size as the original (off axis will then be roughly similar) plus the enclosure must have roughly the same dimensions (baffle effects having identical impact) : Copy the acoustic transfer function of each driver precisely and you will be close.
I predict many here will object strongly ("each driver has its own sound signature.."). Of course the drivers must not have strong break-up in their respective passbands.
There is an experiment by Member Jon Bocani with compression drivers: he concluded that, once minutely equalized, they sounded identical.
He got flamed here, but there was little hard evidence against his case. Furthermore, member MarkBakk posted a German experiment with equalized tweeters, that came to the same conclusion. Earl Geddes is also of the opinion compression drivers, regardless of price, are basically a commodity if they meet certain mininum specs: it is the waveguide that makes most of the difference.
One may have some questions, but it is an experiment I would like to undertake myself. Listening to pass band filtered midranges having the same size, housing, level and acoustic transfer function would be an interesting starting point.
Oops, now I have put my head in the guillotine..
Do the test with intended bandwidth and remember to state what that is with the conclusions 🙂 I mean excursion (associated with with bass frequencies) makes distortion on the mids, depending on the amplifier and possible passive components between amp and the driver(s). Been reading some stuff on how driver impedance depends on voice coil position, excursion, and it seems logical. See the Elsinore thread and Joe Rasmussen posts from past few weeks or so. Also what Purifi has done with their drivers.
As i have understood it even if transfer functions of two similar drivers were equalized to be the same at some particular SPL they would differ at some other. Differences could hide with some test material like sweeps and be more apparent with multitone tests, I think.