Understanding the state of the DSP market

[jeffme]

...when I evaluate a DSP product, what I want to know is:

- What type of DSP filtering does it do? IIR or FIR?
- How many biquads/taps?
- What does the software do? Can you use third party software if you don't like the software they recommend? Are you locked in to their software? Do you need to pay additional license fees (Dirac) or worse, a subscription (anything that uses Audioweaver)?

If they hide this information on their website, I look somewhere else. I need to know exactly what I am buying, and if they don't tell me - I cross it off my list. A bit harsh, I know. But I would expect a website to provide this information.

@Keith_W - many thanks for initiating this great conversation.

In the quote above, are these specs we should be looking for as starting points for discussion of any DSP features in any products? I mean, if those are the baseline questions you're asking, seems like they should be the baseline questions we're all asking. Aren't these the kind of requests ASR readers could/should start asking of manufacturers?

For example, if I had a WiiM streamer with DSP feeding into a Sonance amp with DSP, wouldn't I need to know this info in order to make a better decision on where to apply my filters? Or to compare the DSP built into different amps? Or, if I already had a built-in DSP and was considering an upgrade, wouldn't I want to know what I'm upgrading *from*?

And... totally understood this is before we get to the user error parts of DSP.

 

[Keith_W]

Hi @Keith_W.
I was deepening the market situation of Dirac ART and given the title of the thread I'm wondering why you skipped a bit on it, although it seems to theoretically represent the most versatile and effective DRC technology (Trinnov is a bit more hardware related). I see that there are delays in rollout and some bugs, but I also see incredibly promising results, unrivalled by simple FDW + inversion filters, even with low count speakers.
Beyond reliability/flexibility, do you know any other aspects that manual software/methods are preferable for?
Do you have any comparison of results between Dirac and Acourate for example?

I don't know how Dirac ART works beyond "we use this speaker to cancel modes caused by another speaker". It's a bit too vague to be useful. And Trinnov is the same, except that their scheme sounds very similar to a DBA. So I do not know whether it is a real innovation or not, because it's not as if schemes like these do not already exist. I would need to know in detail exactly what they are doing before I can have an opinion. And I don't have opinions about things I don't know, I only have questions.

As for Dirac vs. Acourate, the difference in a nutshell is: one is like a bread making machine, the other is a collection of kitchen tools - a stand mixer, oven, baking dishes, and so on. The bread making machine directs you to add flour, water, and yeast. It kneads it for you, rests the dough, then bakes it. You will consistently get good bread, but it can't be used for anything else. With kitchen tools, you have to measure out the flour, water, yeast. Then you have to know how much to knead and how long to rest, then set the oven correctly and bake for the correct time, and know when to remove the bread from the oven. There is a learning curve involved, but you can get so much more than baked bread. You can make cakes, scones, pizza dough, cookies ... all sorts of things. And arguably, you can probably make better bread ... but you have to learn how to do it first. And it will never save you time compared to the bread making machine.

So Dirac is great if you want convenience and you are happy with the result that some engineer decided would be best for you. But you will never learn about DSP and you won't be able to adapt it to do something else. Acourate makes you learn the process, but once you understand what it does, it is really flexible. For example, if you want to make a digitally delayed CBT, you can do it. You can edit curves, for e.g. take a measurement of a woofer and a port, and then stitch them together. You can use it to simulate the effect of a correction, and the simulation is extremely close to reality. It can exploit the ability of linphase filters to independently correct phase and amplitude, and if you don't like what it suggests, you can DIY your own solution. You can read about some correction method on ASR, then go an experiment. IMO the curve editing feature of Acourate is unmatched. REW lets you edit curves, but some features are missing (or I haven't found them yet!). Audiolense won't even let you import a .WAV file, let alone edit the curve. And Dirac can't take verification measurements, you have to use REW.

IMO both solutions have their place. Dirac is better for people who want the result but don't want to learn the nuts and bolts of DSP. Acourate is great if you like to tinker.

I don't have results of a direct comparison between Dirac vs. Acourate, and IMO the comparison would be meaningless. With Acourate, the result depends on the user, a bit like how you can't compare bread made with a bread making machine vs. manual baking. If the baker doesn't know what he's doing, the result will be worse. If the baker is skilled, the result will be better. The bread making machine will always turn out the same, consistent result.

 

[Keith_W]

In the quote above, are these specs we should be looking for as starting points for discussion of any DSP features in any products? I mean, if those are the baseline questions you're asking, seems like they should be the baseline questions we're all asking. Aren't these the kind of requests ASR readers could/should start asking of manufacturers?

I CERTAINLY THINK SO!

For example, I mentioned in my first post that with FIR filters, the more taps you have available, the finer the resolution of correction. The other variable is the sampling rate. The taps are spaced evenly across the sampling rate, and that decides how big the "chunks" (bin size) are going to be. So for example:

2048 taps / 96kHz: 96000/2048 = 46.9Hz.
65536 taps / 48kHz: 48000/65536 = 0.73Hz.

Here is an example: the original crossover for my sub (in red) has been cut down to 8192 taps/48kHz (brown) and then 1024 taps/48kHz (green). The loss of resolution is very easy to see.

So now whenever you see a DSP product offer FIR filtering, you want to know how many taps and at what sampling rate because that tells you how useful (or useless) the product is going to be. Let us look at the MiniDSP Flex Eight. The specs page says 2048 taps per input channel at 96kHz, or 4096 taps in total. That would give you the same result as the green line in my simulation above (1024 taps @ 48kHz).

So in the MiniDSP Flex Eight, the IIR biquads do all the heavy lifting. I am not sure if the FIR filters can do very much. It may be possible to increase the FIR resolution in the low freqs by using something like logarithmic taps but I don't think MiniDSP has implemented such technology.

 

[LionIT]

I don't know how Dirac ART works beyond "we use this speaker to cancel modes caused by another speaker". It's a bit too vague to be useful. And Trinnov is the same, except that their scheme sounds very similar to a DBA. So I do not know whether it is a real innovation or not, because it's not as if schemes like these do not already exist. I would need to know in detail exactly what they are doing before I can have an opinion. And I don't have opinions about things I don't know, I only have questions.

As for Dirac vs. Acourate, the difference in a nutshell is: one is like a bread making machine, the other is a collection of kitchen tools - a stand mixer, oven, baking dishes, and so on. The bread making machine directs you to add flour, water, and yeast. It kneads it for you, rests the dough, then bakes it. You will consistently get good bread, but it can't be used for anything else. With kitchen tools, you have to measure out the flour, water, yeast. Then you have to know how much to knead and how long to rest, then set the oven correctly and bake for the correct time, and know when to remove the bread from the oven. There is a learning curve involved, but you can get so much more than baked bread. You can make cakes, scones, pizza dough, cookies ... all sorts of things. And arguably, you can probably make better bread ... but you have to learn how to do it first. And it will never save you time compared to the bread making machine.

So Dirac is great if you want convenience and you are happy with the result that some engineer decided would be best for you. But you will never learn about DSP and you won't be able to adapt it to do something else. Acourate makes you learn the process, but once you understand what it does, it is really flexible. For example, if you want to make a digitally delayed CBT, you can do it. You can edit curves, for e.g. take a measurement of a woofer and a port, and then stitch them together. You can use it to simulate the effect of a correction, and the simulation is extremely close to reality. It can exploit the ability of linphase filters to independently correct phase and amplitude, and if you don't like what it suggests, you can DIY your own solution. You can read about some correction method on ASR, then go an experiment. IMO the curve editing feature of Acourate is unmatched. REW lets you edit curves, but some features are missing (or I haven't found them yet!). Audiolense won't even let you import a .WAV file, let alone edit the curve. And Dirac can't take verification measurements, you have to use REW.

IMO both solutions have their place. Dirac is better for people who want the result but don't want to learn the nuts and bolts of DSP. Acourate is great if you like to tinker.

I don't have results of a direct comparison between Dirac vs. Acourate, and IMO the comparison would be meaningless. With Acourate, the result depends on the user, a bit like how you can't compare bread made with a bread making machine vs. manual baking. If the baker doesn't know what he's doing, the result will be worse. If the baker is skilled, the result will be better. The bread making machine will always turn out the same, consistent result.

Okay, you're focusing on DIY. The title of the thread made me understand the topic a little differently. This is why I was wondering why you did not reserve a more in-depth description of a technology like ART, which on paper allows you to achieve impossible results with usual filtering, even in terms of latency.

 

[levimax]

Okay, you're focusing on DIY. The title of the thread made me understand the topic a little differently. This is why I was wondering why you did not reserve a more in-depth description of a technology like ART, which on paper allows you to achieve impossible results with usual filtering, even in terms of latency.

I sincerely hope that DIRAC can get ART working but so far they have not even been able to get DLBC working for many use cases (like mine) and you can read about it on these boards. ART is orders of magnitude more complicated and I am skeptical that the limitations / uncertainties of in-room measurements will allow ART to work reliably across a large range of use cases. Again I hope I'm wrong but so far ART is pretty much "vaporware" except for some Storm Audio devices which are very expensive.

 

[DWPress]

DLBC works fine on PC. The hardware bundles all have underpowered processors as demonstrated by Keith above.

 

[levimax]

DLBC works fine on PC. The hardware bundles all have underpowered processors as demonstrated by Keith above.

Not on my PC it doesn't. If I run it 5 times I get 5 different corrections and the delays are not even close... like 125 ms when 5 ms is the right answer. If you read over on the "Exclamation Mark Issue" thread https://www.audiosciencereview.com/forum/index.php?threads/dirac-live-exclamation-mark-issue.52443/ you will see it seems to work for some people and not for others. It seems to do a good job with FR but timing is another story for many including me.

 

[DWPress]

Not on my PC it doesn't. If I run it 5 times I get 5 different corrections and the delays are not even close... like 125 ms when 5 ms is the right answer. If you read over on the "Exclamation Mark Issue" thread https://www.audiosciencereview.com/forum/index.php?threads/dirac-live-exclamation-mark-issue.52443/ you will see it seems to work for some people and not for others. It seems to do a good job with FR but timing is another story for many including me.

It works well on Mac, surprisingly I guess. I get consistent delays on mains and subs when I run at least the 5 position sweeps. Of course every time I run the correction software and do sweeps it comes up with a slightly different result but things move around my room, same with REW. I've looked at the thread you linked to but it seemed like it was mostly AVR and miniDSP hardware/software issues. Not a Dirac fan boy by any means otherwise.

 

[Fidji]

My personal ranking of DRC systems [all in one of my setups, so very individual, YMMV], both listening and measurements, bass only:

1) Trinnov Waveforming in DBA like 8 sub configuration
2) Same sub config like above, but DBA manual setup
3) Cylindrical Waveforming (just as test using 4 subs)
4) 6 subs running MSO via 2x Flex HT
5) 2x Flex HT and REW manual sub adjustment
6) Trinnov single sub (4 subs EQ’ed as one unit)
7) DLBC - sonic atrocity with audible Group delay issues [i freely admit it might be my setup/calibration]

Dead Ends:
1) “REL Six Pack” config - 2x3 stacked subs and Audyssey
2) Audyssey with more than 3 subs
3) REL 3D - assigning subs to imdividual speaker groups.
4) Any system with indicidual sub calibration - in my case Anthem ARC for MartinLogan subs.

 

[3ll3d00d]

Dirac is great if you want convenience and you are happy with the result that some engineer decided would be best for you.

Not sure that's really true, every single such product ends up with lengthy threads in which people advise each other on how to get reliable/good results so you need to know what you are doing to some degree a d understand how to take measurements & how that influences the end result. All flavours of dirac and audyssey have had such threads on different sites and current versions of those products are all configurable to varying degrees, it's mostly about the target curve of course because those products don't do crossovers. Trinnov are at the extreme end of this in that the standard recommendation is "get a pro to do it for you" given the range of somewhat obscure options that are available but that might be more a function of the sort of userbase you get when your base product costs something like £17k albeit trinnov does also support active speaker configurations so does have another level of complexity available Vs dirac.

You can compare acourate macros to these as broadly equivalent I think, ie it's a few measurements, a couple of options and you have a filter.

Acourate lets you do a lot more of course and that can be a nice option to have if you want it but I don't think it's essential to getting a good result.

 

[Lion♡]

7) DLBC - sonic atrocity with audible Group delay issues [i freely admit it might be my setup/calibration]

It’s probably not an issue with your setup.

 

[Jumu]

All hardware processors on the market (at the time of writing) use IIR filtering, and some of them have a limited number of FIR taps for mixed-phase processing. You typically get 1024 taps per channel, which is nowhere near enough to implement a high resolution FIR filter. The sole exception is the DEQX Premate 8,

It is an extremely informative post. Thank you.

Do you know what Trinnov uses, IIR or FIR. It has intel i3/i5 chips in them. BTW, I love the storm comment

 

[Sokel]

What's not usually pointed out hard enough is data acquisition. Bad data equals bad results no matter what any of the solutions do afterwards.

And although tons of stuff are written about the "afterwards" the comparison with how to get this data is staggering.
Standards outside the pro world are nowhere and even there are not the norm.

So the usual order goes like this:
- mild interest about RC and stuff or active speaker building
- deep dive to it, long and controversial learning curves, lots of opinions, measurement's frenzy without protocol or DIY ones, etc .
- questioning all the above after experiencing what this is all about listening to a coherent, probably made by heavy educated pros system.
- ditching everything, hiring pros or back to first square doing everything "fresh" but without a way to get those precious reliable, useful data so the loophole goes on.

I think it would be far more useful to try decipher and replicate what pros do even if their gear is a big part of it and their cost is sky-high sometimes (saw a measuring mic array for 10k I think) .
Despite that though, methodology would get people closer.

 

[LionIT]

What's not usually pointed out hard enough is data acquisition. Bad data equals bad results no matter what any of the solutions do afterwards.

And although tons of stuff are written about the "afterwards" the comparison with how to get this data is staggering.
Standards outside the pro world are nowhere and even there are not the norm.

So the usual order goes like this:
- mild interest about RC and stuff or active speaker building
- deep dive to it, long and controversial learning curves, lots of opinions, measurement's frenzy without protocol or DIY ones, etc .
- questioning all the above after experiencing what this is all about listening to a coherent, probably made by heavy educated pros system.
- ditching everything, hiring pros or back to first square doing everything "fresh" but without a way to get those precious reliable, useful data so the loophole goes on.

I think it would be far more useful to try decipher and replicate what pros do even if their gear is a big part of it and their cost is sky-high sometimes (saw a measuring mic array for 10k I think) .
Despite that though, methodology would get people closer.

I agree about relevance of measurements quality.
But, I would add a rhetorical question.
Once you have the most accurate measurement in the world, how do you treat it? What information exactly do you want from it? And above all, where should you want to go then exactly?
Here too there are no standards, as I said a few posts ago, and it is boring given how much science there is in this world.
But somewhere we need to start, and a not so accurate measurement is better then nothing, otherwise we are going to rely on our questionable and perhaps less reliable biological measuring system
Clearly, DSP is just a way to go from A to B here.

 

[Sokel]

I agree about relevance of measurements quality.
But, I would add a rhetorical question.
Once you have the most accurate measurement in the world, how do you treat it? What information exactly do you want from it? And above all, where should you want to go then exactly?
Here too there are no standards, as I said a few posts ago, and it is boring given how much science there is in this world.
But somewhere we need to start, and a not so accurate measurement is better then nothing, otherwise we are going to rely on our questionable and perhaps less reliable biological measuring system
Clearly, DSP is just a way to go from A to B here.

I agree 100% .
A good start and good tools are better than nothing.
I would add big banners too, like "DON'T FILL 30dB DIPS" ,OR "don't stuck filters on top of each other unless you have a very educated look about the combined effects" and so on, the works.

It sounds silly but if one take a tour around threads I just scratch the surface.

And another:
People aching about educated looks but given the very small number of people willing to do it for free as the market also has good places for them I don't see it happen in full. Only hints here and there. I justify it of course, such education has to return its money, it's not free.

So yes, experimenting is the way to go.
And if it's only about playing (making this a hobby) it has glorious avenues to walk through.

 

[OCA]

given the very small number of people willing to do it for free

What I find even more ironic is that an even smaller number of people are willing to use a free DSP tool. It seems that a product only gains respect when there’s a price tag attached. This is likely because only paid products get proper promotion on these platforms, but it’s still noticeable.

 

[Sokel]

What I find even more ironic is that an even smaller number of people are willing to use a free tool. It seems that a product only gains respect when there’s a price tag attached. This is likely because only paid products get proper promotion on these platforms, but it’s still noticeable.

That's unfortunate, yes.
To tell the truth though it seems that it's the vocal scrutinizers who dominate some threads and ask for way more than an automated system offers.
I don't think they are the majority, only louder

And then the confusion, some do HT some do 2-channel some want them both and some want them all, RC, x-overs and speaker correction.

Additionally, a price tag bounds people to these tools. And they rightfully want some decent return for their money. That creates commitment.

 

[LionIT]

That's unfortunate, yes.
To tell the truth though it seems that it's the vocal scrutinizers who dominate some threads and ask for way more than an automated system offers.
I don't think they are the majority, only louder

And then the confusion, some do HT some do 2-channel some want them both and some want them all, RC, x-overs and speaker correction.

Additionally, a price tag bounds people to these tools. And they rightfully want some decent return for their money. That creates commitment.

For what my experience is, Audiolense does all of this and mostly automatically. Never seen users unsatisfied or with acoustic problems that are sw related. It's not free, unless you take it from the usual sources, but it doesn't cost much either considering what it does, how it does and the time it saves you. It costs as much as a single low-end speaker.
The only problem is that, like Acourate, it is developed by a single person (now).
Apart from this, technically I see no reason not to use it, nor do I see great possibilities to do better (perceptually speaking) by choosing the DIY way.
Obviously, glad to be enlightened by someone in this regard
I think basically the point is like the choice between buying a speaker and building one. If I build it, it is more for passion and fun, or perhaps for budget problems.
I do not think there is any wonder therefore about the use of the DSP.

In fact, now that I think about it, an interesting topic on the DSP that is a bit in the shade, is the crossover. I am trying my hand with the digital crossover on my R3s, but I am realizing that online there are many beautiful words and guides on digital crossovers, but few that remind you That if you try to do it yourself at home, without an anechoic room, you risk doing more damage than anything else, without even realizing it. Mainly because of driver's directivity and the way it interacts with the other driver' directivity and with the room (you can produce bumps and dips in the sound power curve, and we now what this means).
So, you can educate yourself as much as you want on the DSP and be able to integrate two drivers in a perfectly linear way at mic position, but if you forget to deal with directivity you may have specialized in this task for nothing.

 

[JanRSmit]

The only problem is that, like Acourate, it is developed by a single person (now).

Why is that a problem for you?

 

[JanRSmit]

What's not usually pointed out hard enough is data acquisition. Bad data equals bad results no matter what any of the solutions do afterwards.

And although tons of stuff are written about the "afterwards" the comparison with how to get this data is staggering.
Standards outside the pro world are nowhere and even there are not the norm.

So the usual order goes like this:
- mild interest about RC and stuff or active speaker building
- deep dive to it, long and controversial learning curves, lots of opinions, measurement's frenzy without protocol or DIY ones, etc .
- questioning all the above after experiencing what this is all about listening to a coherent, probably made by heavy educated pros system.
- ditching everything, hiring pros or back to first square doing everything "fresh" but without a way to get those precious reliable, useful data so the loophole goes on.

I think it would be far more useful to try decipher and replicate what pros do even if their gear is a big part of it and their cost is sky-high sometimes (saw a measuring mic array for 10k I think) .
Despite that though, methodology would get people closer.

What is bad data? Or when is data bad?

With pros you mean speaker manufacturers?
IF so why should i decipher and replicate what these pros do, given the huge differences in speakers?