Neural networks are not very complicated to program or make. It's really the training that is complicated. I'm more interested in gamma re-synthesis personally.

-D

I've owned my Hartmann Neuron since it came out and have always been in love with this synth... a true one-of-a-kind

As many here know, I'm fairly vocal and active member around here and speak my mind heh ... I'll say this about the Neuron and the ideas it represents:

It is THE most innovative synth in the past few decades!!!

My only wish is that you guys could have continued your work and I would have more of your instruments in my studio right now.

By all means, PLEASE keep me (and us) informed of your future work.

Highest regards,

Andrews

Whats that ?

I dont have a doubt in my mind that Neuron is the most sophisticated commercial synth ever made. And it will be , for a veryyyyyy long time.

OK, I will take your word for it but you will allow me to have my doubts about this:

Regarding this:

I have sent them several emails, not only about the inner guts of the Neuron and they never got into trouble of saying "Thanks for the email". Do you think this is appropriate? About 3-4 months ago I sent an email to Prism Sound to request some specific information about an Orpheus and guess what? They called me on the phone as soon as they got my email. They were very polite, they asked specific questions about my setup and what I wanted to achieve, they offered a solution, they suggested sending me an Orpheus AT NO COST (if I didn't like I could sent it back, no questions asked), they have called their dealer in Athens and told them to contact me immediately, they found me a studio where I could hear an Orpheus for myself... The Orpheus costs something like 3500 euros now and the Neuron costed 5000 euros back in 2004. That was all from a company who was basically a newcomer in the world of synths and a boutique company if I may add. So, if I am to fork out 5000 euros I expect nothing less than excellent support.

I can mention several other facts that come to mind right now. You can just search their forum.

Why did it strike you as weird that I have called these components weird? heh let me clarify: weird is the fact that they have chosen to basically construct a PC for such a special technology. You said they had their good reason for choosing this over DSP processors - again I will take your word for it, although I am 100% sure that they could have coax far more potential from a dedicated DSP. They have chosen to build a PC, so I look inside and what do I see? the usual suspects. Hard disk, fan, memory, the Pentium III... don't you expect to see something similar when you open a PC?

The one I played was built very well. The fact that, as you say, early units were poorly built and had to be re-assembled only begs for another question: Why? for 5000 euros? A fan that emulates a lawn-machine? Do you want me to install my application EvoMusic which runs evolutionary algorithms for electronic music synthesis and evaluation, in a mediocre PC, in a very good quality box with fancy controllers and sell it to you for 5000 euros? (It was written in Visual C++ 6.0 for Windows but I can easily port it for Linux so that I don't have to pay anything to Microsoft. heh) Do you see what I mean?

Regarding the fancy mushroom controllers: I liked these, they seemed a tad fragile though. One ergonomic mistake, (at least in my book), was that they have placed one such controller where someone might expect pitch/mod wheels. It just feels unnatural no matter what. You might argue that transposing symphonic orchestra samples with a pitch-bend wheel is also unnatural but it was more so for that little mushroom at the left-most side.

I agree 1000%.

Ah - but did you open it up? The one I saw from the inside had a spot-welded case.

The rest of the control surface is expensive. Getting your R&D back is expensive. The early adopters pay for this.

A better option might've been the same analog joysticks they used for the Playstation and XBOX - minus the centering mechanism. Those can stand some abuse.

It sounds like it would've been more viable to simply keep the case and design thereof a secret and continue developing on standard hardware - and only when that was stable, start publishing pictures. Pictures of new synths = blood for sharks.

Hey OD, thanks for taking the time to talk with me about all that. A ton of great info, much appreciated

-Andrews

Anytime man!thumbsup

I need your help regarding the AnaMod, you still have one, right?

I would say no, probably not. No answer is never a good answer - but I didn't work there so I wouldn't know how they handled tech support. I am surprised to hear this though as I know of many instances where Axel Hartmann himself got in touch with users and knowing him I would say he is the last person to refuse to answer any questions.

You should expect no less, that's for sure. Short them telling you all their trade secrets, that is :-)

No, like I said, at the time that architecture was the best possible solution for their application (the actual parts were not, but that is a different story).

That is the right question and no doubt part of the problem why Hartmann had to file for insolvency. They really made some bad decisions wrt. manufacturing, I agree 100% with you on this.

No not really. However if EvoMusic would be a new synthesis technology available only as a $5k synth and its interaction as a live instrument would demand for this form factor - yes, I might consider purchasing it if I liked the results and if it would spark my creativity. But if you really wanted to follow your line of thought consequently you would have to criticize users of ProTools, the KYMA system etc. for the same reason: their core hardware is built around DSPs that are even cheaper and more commonplace than the components used in the Neuron, and if you get one of these systems for $5k it would be a real bargain. The sum sometimes is greater than its parts because there is so much more involved than just assembling the raw components.

Best wishes
--smb

No, to be honest I didn't. Basically, as far as I can remember, you couldn't open it up as that would void the 1-year warranty. I remember that there were plenty of users, that wanted to change the fan... Damn that fan...

I know what you mean, but I still believe it was TOO expensive.

It's basically a type of linear adaptive IIR filter with a restricted feedback path which makes it easier to adapt. It's like an adaptive FIR filter whose memory depth and resolution you can control with feedback and adaptation rather than needing to modify filter structure.

-D

I will pretend that I understood what the @#$%^ you are talking about ... heh ... still sounds very impresive to my ears.

I love resynthesis, I had loads of fun with Chameleon 5000, Cube 2, Tetra 3 and Linplug Octopus.

At the time that the Neuron came out, I doubt that a single DSP would have been able to match the power of a Pentium III. The TigerSHARCs may have been out at that time, but these are not easy to program. A multi-DSP solution would have worked, but designing and programming such an architecture is not trivial.

The big issue with porting an algorithm over to DSP chips is the development time. The cost of the chips is becoming inconsequential - for any hardware box, you will be spending far more on the knobs, pots, etc., than on any DSPs or microcontrollers. However, most DSPs don't run high-level languages that efficiently. Instead of being able to use your clean C++ code, you need to port things into assembly, which takes time and is a great way to introduce tons of bugs. Plus, a DSP is only good at crunching numbers, and is horrible at things like knowing when notes are triggered and when they end, so you usually need to have a separate microprocessor controlling the DSPs.

If you are putting out a commodity product, like a car stereo system, then obviously DSPs are the way to go, as the development costs will be offset by the cheapness of the parts when multiplied by a huge number of items built. Music gear is becoming less and less of a commodity. If you are only going to sell a few thousand of a particular piece of gear, and it is a high-end item in the first place, spending more on the silicon may end up being cheaper than going to a DSP solution that requires far more development time.

Yes that is exactly one of the reasons why it was decided to use a general purpose CPU based system. Some other considerations were connectivity (ethernet, USB support), using audio hardware with existing drivers, MIDI support, memory and HD considerations and most importantly being able to use the same code base for the NeuronVS which runs as a plug in on a host system.

We're not just talking about a set of filters or even the usual size ANN here - the Neuron DSP code is huge, porting it to a DSP and debugging it would have easily taken an extra year or so. Also, as far as the price tag is concerned, the cheap DSPs are all fixed point - it would have been nearly impossible to port a highly recursive, inherently floating point algorithm such as the Neuron synthesis to a fixed point DSP without losing considerably more than half of the cycles to re-scaling.

If Hartmann had designed the Neuron around a floating point DSP that had all the required features they would have ended up with a dramatically higher price tag than they did with a general purpose CPU approach.

Best wishes
--smb

Always wanted one, any new development would be a great interest - one of a kind, if the re-devloped the engine I'd be all over it.

good grief... just imagine the moaning if the cost of the unit was even higher.

If they ever come out with a V.2, I'll be all over it. Unfortunately, I seem to be attracted to synths that will most likely never see the light of day... Stromberg anyone?

Close enough for now:

Waldorf Music | Q+

I already have a Q... the analog filters on the + would be sick though.

it would be cool if they made a plug-in version for pc/mac like the vs (i tried to buy one on ebay a couple times but declined because its just not stable). i'm sure they could sell it for $500 like metasynth or omnisphere. when it comes to copy protection thats where i guess this becomes a pipe dream. making a plastic controller dongle is expensive and i suggest letting novation or whatever make those to reduce overhead costs. $5000 synths just don't fly not since like the 80s. hartmann always seem to be working on stuff but i just wish a new synth plug would be released that would be totally awesome.

I partly agree, but I'd prefer some hardware (even if just a processor and drive like the Panacara) because the hartmann uses SO much processing and needs drive space.

It'd be REALLY cool if they treated it in a similar way to Kyma's model

-Andrews

Not sure what you mean --- Hartmann Music went out of business in 2005 (or was it 2006?).

Prosoniq, on the other hand, have hinted that they are actually working on a new Neuron-based synth. They haven't said wether hard- or software.


heh

Yeah, that would be very cool. Maybe a chassis for expansion cards, maybe graphics cards for processing. Imagine 8 of the fastest current graphics boards doing audio. Hello, mr. petaflop, whaaazzzzuup heh

In general I'd like it to be less *synth* and more *production tool*, like the Kyma. But rest assured that I'll be whispering that sentence into Prosoniq's ears on a daily basis ;-)

If prosoniq own the rights to the neuron. Then how about just fixing the VS memory leaks and re issuing the VS. As a VS owner I would pay good money to have this unit running without crashing. Hell I would even buy it all over again just for the bug fixes.

Just listen to track 15 - Violine
Winter | Polybonk.com

2 alternating single notes on a keyboard and about 1cm of movement on the stick! Never in my life have I found a synth that sounds anything like the neuron.

Wish I could afford one of these

YouTube - Hartmann Neuron - audio demo - textures and pads

very nice

---

my 10pence:

I think if you wanted to emulate this some how, you would have to use supercollider... maybe Max/Msp. I'm not sure if its possible, but thats my best guess.

Hi Denis, I am currently focusing on Neuron VS.

I am wondering can you share which models are non-datareduced?

I assume 'model complexity' is about number of neurons and/or number of neuron layers?

I am also looking for guidance on what kind of source audio is "good" (and also bad). I gather there should be no noise if possible. I have no clue how long or short the audio should be, or can be. Minimums and maximums, best lengths? I am unable to find guidance anywhere.

Thanks very much in advance.

Just read a bit on hartmann's design history page and he mentions stephan sprenger as the dsp guy .
Wasn't it stephan Bernsee , or are these the same people ??

Quote

My enthusiasm grew stronger still when I met Stephan Sprenger — a specialist in the use of neural networks in music applications — at Musikmesse in 2000. I explained to Stephan my idea of analysing known sounds with the help of adaptive algorithms to understand them and having parameters that allow for sound manipulation based on human levels of perception.

Unquote

Edit: yep they are one and the same person

Same person.

Okay, I'll play – but bear with me, it's been a while ;-)

IIRC, Models 0,1,2,3,4 and 5 are non-data reduced, and 9 is somewhere in-between. You'll be able to tell the difference immediately though, the sound is VERY obviously lower res on the data reduced models.

I can't say what the complexity does exactly, but as the parameters keep doing what they're doing I don't think the neural network topology is changed.

Sounds that have one distinct character work best (this is similar to the "no noise" assumption). Shorter vs longer doesn't have a huge impact in the real world, though I'd say that longer samples would theoretically work less good.

In general I'd say just go and play with it

Thank you Denis! Very Helpful!