u-he rePro in the works

[Numanoid]

[VariKusBrainZ]

It isn't just a software or software optimisation problem though. FPGAs and ASICs are custom hardware too. In the case of the former the gates can be reprogrammed and thus the device repurposed to a new task. Different looking silicon all of a sudden. Maybe you're tired of processing reverb. Maybe you want to process something else now. With your i7 you're frozen in silicon after it leaves the factory. Can't change it. With FPGAs you can.

I'd be very surprised if any company reprogram anFPGA real-time and in field-shipped units or even with "firmware" updates. They'd need to add hardware and software subsystems to reprogram the FPGA in the shipping unit which would add to the overall device cost and software maintenance overhead. I suppose it's possible but I'd wager the amount of R&D to get it right is better invested in a MkII product. So in that sense I don't see how an FPGA is any less frozen than an ASIC in a shipping product from the customer's point of view.

In the case of the Bricasti M7 they use Analog Devices dual core DSPs which are frozen silicon processors just like the CPU in your phone/tablet/PC. Here's an example of one of their higher end Blackfin models: http://www.analog.com/en/products/proce ... bf606.html
It's likely similar to the one (of six) Bricasti uses in each M7. No FPGA there and it runs at 800MHz. The difference is specialization and the software layer to interface with it.

Like the Virtex chip that was going to revolutionise modern computing back when it was announced

[Urs]

indeed!

It's fun how hardware guys come up with scientifically backed up claims that their hardware is somehow superior for the special case they have. Only, they of course don't unveil what that special case might be. They say we need punzillaflops of fused multiply adds. Fair enough, but do they? In some cases where they say "a desktop CPU can never do this" it's a matter of five days to create an algorithm thats suspiciously similar sounding. That's how long it took to write the core of Protoverb. I also think that layering 10 Schroeder style reverbs with a total of 30 seconds of delay is a piece of cake, and might get you there beyond boundaries of perception.

And then, every now and then something miraculous happens. Like, waveshapers that used to cost 30 operations to approximate so-so suddenly cost 25 operations to calculate accurately. And suddenly a whole different set of methods can deployed to model something much more accurately at half the expense.

Furthermore it's six years since development of Diva started. Meanwhile I have dived into circuit simulation and supplementing subject areas at an extent that equals the knowledge transfer of acquiring a university degree. If there is one thing for sure, then it is this:

The effects, faults and mathematics of electronic components are understood sufficiently well enough to be modeled discretely beyond the boundaries of human perception.

What it means is, it can be done. Period. However, a lot of areas have not been explored in as much depth as I would have loved. Developers are very secretive about their choice of algorithm to achieve the same thing. We lack in depth studies as to which method excells in which way. We furthermore lack in depth studies (conducted by boffins, not gear nuts on Youtube) as to how different two individual analogue synths really sound, and why. Last but not least we lack knowledge about "how accurate is accurate enough?" - Each of these fields could lead to faster, better and more convincing simulations of said analogue circuits.

So, guys at u-he are about to arrive at that spot. Over the years we had interns explore various known methods to do the same thing, and we've added novel methods and algorithms on our own. I think we can sincerely claim, we can pretty much compare a comprehensive set of methods to model analogue filter circuits, and we can do so in any level of detail, i.e. with better or lesser approximations of the parts involved. We can also model the tolerances of parts and thus reproduce individual differences between specimen. We're equipped with that knowledge, and we were able to gather it over the years all by ourselves.

The one part that we haven't done yet is measure the boundaries of perception among the target audience. That is, we would like to know how siginificant the accuracy of the model and the method needs to be to fulfill the task. This would have a direct effect on everything that's related to the CPU consumption. So yes, this is pretty much what it is about. It's not a question of "does it run or not?" anymore, it's a question of "how fast can we go". And believe me, as the creator of Diva I know every aspect of why users want faster algorithms. Now, this knowledge can not be aquired from books. It needs to be empirically harvested from you guys out there who use the stuff.

So this is what RePro is. A small research project on human perception. It may have an effect on our design decisions and choice of algorithms in virtually any product. It's not just "yet another analogue emulation". It's an awesome concept designed to make Zebra, Diva and possibly every other u-he product better sounding in the most efficient way.

[nevernamed]

What makes a convincing emulation is precisely the thing to focus on. Arguably you're already pushing the limits of perception/cognition with Diva. It isn't at 5% CPU but that is neither here nor there and time is on your side.

There are cases where hardware is more appropriate or efficient etc. That isn't to say similar results couldn't be obtained in software; efficiency and power utilisation notwithstanding.

How fast can we go is the point yes. Would you decline more CPU if it were available? Probably not. More is useful more often than not.

The perception thing won't be objective strictly but I think this is a matter of degree. You're on the right thought there.

Are there any plans for Diva in hardware? There is something to be said for tactile interfaces and exploration.

[Urs]

Diva hardware is tempting, but we can't afford to bet a years revenue on an experiment.

I might reconsider once the current line of work is done. Last year we fixed things, this year we move things forward. As exciting as this is, we need to keep level headed.

[JonSolo]

Diva hardware is tempting, but we can't afford to bet a years revenue on an experiment.

I might reconsider once the current line of work is done. Last year we fixed things, this year we move things forward. As exciting as this is, we need to keep level headed.

Good. Do what you do best...and you guys are awesome at what you do. Looking forward to see how this all develops.

[.jon]

What makes a convincing emulation is precisely the thing to focus on.

Emulations were born in a time when real analog synths were rare and expensive, unobtanium for most people, and those who had them cursed at the decades old tech. Those days are long gone. I can walk into a mainstream consumer electronic retail store and buy a new Moog, Arp or Prophet. Or I can order modules to build a custom synth with parts from Roland, Moog, Waldorf and DSI plus a ton of more innovative designs. These cost more than a plugin, but aren't worthless in 10 years, and are way more inspiring on top of sounding better.

Now what if all those man-years of trying to create passable approximations of how old circuits sound would have been spent on researching PM? UX and UI design to better access additive synthesis? Exploring granular synthesis? What else is out there? Will we ever get to know, if all the effort is spent on reselling the same old "realistic vintage analog emulation!!111" over and over again?

What I mean is that if we accept that subtractive is the shit and bees knees of all synthesis, then it was already done well 50 years ago, and doing a passable emulation of it on the PC isn't going forwards, it's going backwards while never reaching the goal.

[pdxindy]

What makes a convincing emulation is precisely the thing to focus on.

Emulations were born in a time when real analog synths were rare and expensive, unobtanium for most people, and those who had them cursed at the decades old tech. Those days are long gone. I can walk into a mainstream consumer electronic retail store and buy a new Moog, Arp or Prophet. Or I can order modules to build a custom synth with parts from Roland, Moog, Waldorf and DSI plus a ton of more innovative designs. These cost more than a plugin, but aren't worthless in 10 years, and are way more inspiring on top of sounding better.

Now what if all those man-years of trying to create passable approximations of how old circuits sound would have been spent on researching PM? UX and UI design to better access additive synthesis? Exploring granular synthesis? What else is out there? Will we ever get to know, if all the effort is spent on reselling the same old "realistic vintage analog emulation!!111" over and over again?

I love Diva... it sounds great! and I can have 16 voices and multiple instances. How much would multiple instances of 16 analog voices cost?

What I mean is that if we accept that subtractive is the shit and bees knees of all synthesis, then it was already done well 50 years ago, and doing a passable emulation of it on the PC isn't going forwards, it's going backwards while never reaching the goal.

[Urs]

Now what if all those man-years of trying to create passable approximations of how old circuits sound would have been spent on researching PM?

What if the laws of analogue circuits can be applied to other fields such as Physical Modeling, Phase Modulation (not sure which one you meant by PM) and what have you. What if this whole zero delay feedback business was also the foundation of better FM algorithms (it is!) and, say, better resonators for whatever physical model (it is!) and even better dynamics processors (it is!). What if the insight of what we're dealing with has a fundamental effect on how we create novel algorithms?

Also, you said yourself that softsynths are far from sounding like analogue synths (most probably are...). If that was true, shouldn't even more effort go into this? Because even if those few analogue hardware synths are somewhat affordable today, they're far from practical in a modern, mobile or distributed production environment. Softsynths, being data, do not just have disadvantages, they're not just lesser than hardware. I reckon in contemporary production they are more common than whatever grounds hardware synths will ever regain. Making good sounding software is IMHO more important than whatever renaissance hardware goes through.

[knowix]

Wasn't .jon celebrating the end of the plugin era recently? At this stage I never trust anyone who thinks plugin devs are wasting their time improving emulations. Is .jon aware of how Urs's work with Diva has improved the sound of the filters in Bazille and Hive?

Plus, as I've suggested before, people like Urs and Andy from Cytomic probably have a better grasp of the circuitry of analog goodness than most audio hardware devs these days (*cough* Rhythm Wolf *cough*). Who's to say they won't come up with something that's more analog than analog. Imagine an analog circuit that requires hundreds of precisely specified parts, that would be difficult to mass produce. Or a circuit that only works at a certain temperature band.

As regards what Urs said about people's perceptions of sounds. I think there's a huge likelihood that people can perceive things that current scientific knowledge says they shouldn't be able to hear. I also think that what appears on an oscilloscope/spectrascope doesn't necessarily reflect what happens when a sound bounces around the ears, or the surrounding room.

[trimph1]

Who the beeeep!!!! cares? Plugins and hardware are just as good as each other. I remember when it was all hardware...the electric bills were not fun!!!! And if'n you think you could get away with it now? HAH!!!!

[trimph1]

Same here

[kmonkey]

One need to be complete idiot to believe that Bricasti reverb and processing inside (as much as good it is and no doubt it is good) is 30 times more powerful then Intel i7.

That is like complete nonsense. Yeah i know someone will jump with "under special case" stupidity and all that...

That is complete bu****** reserved for (i guess) their users. Of course that claim is born from person behind the same company which created that reverb. Of course. To me it's like a cheap attempt to make his hardware device somehow superior to everything else in the market. Again complete nonsense.

If anything is even remotely true about that, even it is 5% true then i can safely know that if he need power of 30 i7 to generate reverb sound then he is very bad DSP developer...that's for sure.


Regarding DSP and sofware synths. I don't even bother. They are good enough to me. I still prefer some of my hardware (even digital) synths over software but overall when i look at it - i use everything which sounds good to me no matter if it is hw or SW.

Some years ago when i was reading about supercomputers and chess and all that fancy stuff - i was under that impression that it's not about raw power and having rooms of super computing to compete with one person in chess session. Me think's it is about algorithm. That's all.

Once upon a time someone will create algorithm which will compete and outperform human chessmaster with way way less resources needed because algorithm will be more advanced - not the raw power.

Same will (hopefully) happen (for some things it already happened) with these plugins...U-he is on the good path i hope

[AnX]

it's a matter of five days to create an algorithm thats suspiciously similar sounding.

Maybe take 10 days, and make it sound identical?

[chk071]

Emulations were born in a time when real analog synths were rare and expensive, unobtanium for most people, and those who had them cursed at the decades old tech. Those days are long gone.

Interesting. I thought emulations were born out of the fact that you have many advantageous when making stuff digital, so that unstable oscillators, monophony, unreliable technics, being able to only use one instance and all that could be ruled out, and combined with the ability to add lots of stuff which would cost a fortune if you would put it in the real thing. And those days are long gone? Interesting too, because over the years, i felt like the amount of soft synths actually increased, not decreased.