Softsynth version of famous modern VAs?

[BOBA JFET]

 

Originally posted by teoman Otherwise all hardware VA manufacturers would make their synths on ordinary PC motherboards plus a good sound card, but they don't...

 

 

Why would they do that?

[keyman_sam]

I tried Vaz(v2.1 i think) long time ago, but didnt like the sound. The interface looks very simple with drop-downs for assigning stuff, but i have no complains with using patch cords ala G2, Reaktor style.

 

 

Maybe someday, i'd use Reaktor to recreate all the modules available in G2, and whadya know, i have a nord in my puter!

 

[The Audacity Works]

According to Teoman's logic, when Apple switches to Intel processors, Logic will sound different.

 

Who knows; maybe it will?

 

Nah-- No it won't.

[J3RK]

Originally posted by teoman Eight separate chip sets for 8 different sounds on the Ion ! So Alesis engineers must be knowing something that we don't know. Saying that it sounds almost the same on a PC is a bit oversimplification. For the Virus powercore issue, if that powercore card carries the same DSP chips, of course it must sound very close to the hardware Virus. Nothing surprising about this, but I think my original argument was a bit different. If you run a soft synth on a PC and on a dedicated hardware they will sound slightly different. Otherwise all hardware VA manufacturers would make their synths on ordinary PC motherboards plus a good sound card, but they don't...Using the same chipset with the hardware equivalent on a PCI card like in the case of Virus powercore, Yamaha PLG150AN, etc. will definitely give almost equivalent results.

 

The reason they use dedicated DSPs is that they are the most efficient and cost-effective way to do the necessary calculations. It has nothing to do with audio quality. Purely computational. They could use a PC motherboard if they wanted, but it would be a lot more expensive to produce, it would be less reliable, and there is a lot of overhead involved running an OS. The Neuron and the Oasys use a PC setup. Mathematics determines how good the sound being produced is (on a digital synth,) and converters are responsible for the quality of the output.

[mytee2.0]

Originally posted by audacity works According to Teoman's logic, when Apple switches to Intel processors, Logic will sound different. Who knows; maybe it will? Nah-- No it won't.

 

those 1's and 0's will be a lot warmer

[Teoman]

 

Originally posted by BOBA JFET Please explain how the same instructions will ever yield different results, teoman.

 

 

Yes, the same instructions, but different execution speeds. Does Windows XP work exactly the same on Pentium 2 and Pentium 4? It's the same code, but different performances...If my OB-12 had more DSP chips, I wouldn't face with clipping issues in unison mode and probably a better sound. As far as PC hardware goes, quality mother boards, quality sound cards are reliable too. They choose their own hardware design plus they develop their own operating system (instead of Windows XP) and face with all these bug issues, but they obviously choose this path. It is not true to think software independent of hardware. Yes, many software codes are hardware independent, but performances may differ.

[droolmaster0]

 

Originally posted by teoman Yes, the same instructions, but different execution speeds. Does Windows XP work exactly the same on Pentium 2 and Pentium 4? It's the same code, but different performances...If my OB-12 had more DSP chips, I wouldn't face with clipping issues in unison mode and probably a better sound. As far as PC hardware goes, quality mother boards, quality sound cards are reliable too. They choose their own hardware design plus they develop their own operating system (instead of Windows XP) and face with all these bug issues, but they obviously choose this path. It is not true to think software independent of hardware. Yes, many software codes are hardware independent, but performances may differ.

 

 

This is just so silly. software isn't independent of hardware, of course. It needs hardware to run on. But the same numbers playing through equivalent converters will sound the same. Examples exist of both computer software and hardware/software versions of the same synths, and they predictably sound the same.

 

I just can't believe that this myth that hardware sounds better because of these magical reasons still exists. performance may differ sure - so you might get more voices on this system than that. But that's not what we're talking about.

[Teoman]

Ok, I will stop here... we are at different frequencies, no more comments from me

[BOBA JFET]

 

Originally posted by teoman Yes, the same instructions, but different execution speeds. Does Windows XP work exactly the same on Pentium 2 and Pentium 4? It's the same code, but different performances...If my OB-12 had more DSP chips, I wouldn't face with clipping issues in unison mode and probably a better sound. As far as PC hardware goes, quality mother boards, quality sound cards are reliable too. They choose their own hardware design plus they develop their own operating system (instead of Windows XP) and face with all these bug issues, but they obviously choose this path. It is not true to think software independent of hardware. Yes, many software codes are hardware independent, but performances may differ.

 

 

If what you were saying were true, a softsynth would sound better running on a computer with a Pentium IV than on one with a Pentium II. I can tell you right now this is not the case. All other factors equal, a faster processor will not affect sound quality, only the number of voices, instances, or in some cases the ability to run the softsynth at all.

 

In the case of your OB12, an underpowered DSP will not make unison mode clip. An underpowered DSP may however cause the designers to make compromises in their algorithms, like a less-than-ideal unison mode.

[DrJustice]

Since the possibility of differences between processors has been touched upon, I'll trow in a bit on that:

 

Different DSPs and general purpose CPUs can have different word lengths (the number of bits that make up a number). They may also be able to operate on different number formats, mainly one or both of fixed and floating point.

 

E.g.:

 

A 56k series DSP has 24 bit words and a 56 bit accumulator and can do fixed point arithmetic. A typical GP CPU (pentium'ish) may have 32 bit integers (can be used for fixed point) and perhaps 32, 64 and 80 bit floating point. A SHARC DSP has 32 bit words and an 80 bit accumulator for fixed point, and it has 32 and 40 bit floating point.

 

When the same math operations are performed on these, the precision of the operands and the results are different. In a synth or fx program there will typically be many stages of processing, some of which are recursive (i.e. having feedback).

 

Without going into too great detail, for each stage of processing and each recursion a little bit of precision (typically 0.5 bits...) is lost when the results are being rounded or truncated. Half a bit amounts to ca. 3dB of S/N or dynamics.

 

In real life a synth may have several tens of computational stages in which the loss of precision (or error) accumulates. In the case of different word lengths and number formats these errors cause the results to deviate even further from the ideal (and that of the other CPU's).

 

The effects of these differences and errors can make themselves audible. Sometimes it may be quite noticeable; e.g. recursive filters at low frequencies, say an EQ used in the bass range, are very sensitive to the word lengths used.

 

There is a lot more to be said about all of this, of course. Things aren't always as simple as they may seem at first...

 

myteeGTi write: those 1's and 0's will be a lot warmer

Very simplified, one might say that more bits give more warmth

 

DJ

--

[MorePaul]

now that is an informative and interesting post

 

say -- you're that guy that has mentioned some dev work on that soundart box right?

 

so what's the deal with that?

 

I noticed they have a mkii coming, but it is overdue? (gee, like that *never* happens other places )

[lasselu]

Originally posted by MorePaul so what's the deal with that? I noticed they have a mkii coming, but it is overdue? (gee, like that *never* happens other places )

They've shut down, sadly enough...so no Mk II...

The website is still up and as I understand it you can still buy whatever units they've got left...

I have the Mk I...I haven't developed anything for it as yet...just fiddled with the examples, but it's a great unit...worth it's weight in gold for the Infiltrator skin alone...

[DrJustice]

Paul,

 

yes I've been dabbling some (and still am) with the Chameleon. There's a community of developers and new skins are being made (se below). Unfortunately the market wasn't quite there for the Mk II, so It's been put on ice for now.

 

 

lasselu,

 

They've not quite shut down, more like hibernating. Hence the continued operation of the website. There's been some new independent skins lately over at chamelon.synth.net .

 

There's also recently been an update of Infiltrator (around 15 minutes ago...) to v. 1.02. Nothing major, basically a maintenance update with a few enhancements.

 

DJ

 

P.S. Sorry for the OT keyman_sam - if we get into more Chameleon stuff we'll make a thread.

--

[MorePaul]

I'd like to hear more about it

 

maybe you guys can do a little FYI thread on it

 

I'd certainly appreciate it

 

and it'd clear this for keyman sam

 

Sorry abt that bro!

 

end hijack

 

[keyman_sam]

No problem.

 

This thread has been pretty informative and interesting......

[Teoman]

Very good explanation Dr. Justice.!!! Things are not as simple as they seem.

[BOBA JFET]

Originally posted by DrJustice Since the possibility of differences between processors has been touched upon, I'll trow in a bit on that: Different DSPs and general purpose CPUs can have different word lengths (the number of bits that make up a number). They may also be able to operate on different number formats, mainly one or both of fixed and floating point. E.g.: A 56k series DSP has 24 bit words and a 56 bit accumulator and can do fixed point arithmetic. A typical GP CPU (pentium'ish) may have 32 bit integers (can be used for fixed point) and perhaps 32, 64 and 80 bit floating point. A SHARC DSP has 32 bit words and an 80 bit accumulator for fixed point, and it has 32 and 40 bit floating point. When the same math operations are performed on these, the precision of the operands and the results are different. In a synth or fx program there will typically be many stages of processing, some of which are recursive (i.e. having feedback). Without going into too great detail, for each stage of processing and each recursion a little bit of precision (typically 0.5 bits...) is lost when the results are being rounded or truncated. Half a bit amounts to ca. 3dB of S/N or dynamics. In real life a synth may have several tens of computational stages in which the loss of precision (or error) accumulates. In the case of different word lengths and number formats these errors cause the results to deviate even further from the ideal (and that of the other CPU's). The effects of these differences and errors can make themselves audible. Sometimes it may be quite noticeable; e.g. recursive filters at low frequencies, say an EQ used in the bass range, are very sensitive to the word lengths used. There is a lot more to be said about all of this, of course. Things aren't always as simple as they may seem at first... Very simplified, one might say that more bits give more warmth DJ --

 

Okay, but what does all of that mean in practical terms when running on a SHARC vs. a Pentium IV, for example?

[J3RK]

Not a lot. Yes, there may be slight rounding differences if say, integers are used instead of floating point, but I don't think it's going to be particularly audible if similar functions, and similar outputs are used. Also, I would think that if the software was carefully ported from one platform to another, and not just dumped straight across (assuming that were possible in the first place, (this is all theoretical,)) there would be little if any difference. (other than performance perhaps) This brings up the fact that if one platform performs better, the programmers could take advantage of that, and make the synth sound better. (but then we aren't talking about a 1 to 1 comparison anymore)

[BOBA JFET]

The main thing I'm curious about is if x86 is capable of doing more than 32-bit fixed point math in software. I know that in x86 assemby you can define the registers just about any way you'd like, and that EAX is an accumulator register, but I don't know enough about that to make any hard statements. The greater precision fixed point math of a DSP could be significant.

 

PS - teoman still doesn't know what he's talking about even if he does feel a little bit vindicated right now.

[lasselu]

Originally posted by DrJustice There's also recently been an update of Infiltrator (around 15 minutes ago...) to v. 1.02. Nothing major, basically a maintenance update with a few enhancements.

Cool...

 

Aw crap, I get a "File Not Found" when I try to download it...

[Teoman]

Originally posted by BOBA JFET The main thing I'm curious about is if x86 is capable of doing more than 32-bit fixed point math in software. I know that in x86 assemby you can define the registers just about any way you'd like, and that EAX is an accumulator register, but I don't know enough about that to make any hard statements. The greater precision fixed point math of a DSP could be significant. PS - teoman still doesn't know what he's talking about even if he does feel a little bit vindicated right now.

 

Believe me, I am right ! even simple latency differences in a sound card makes a difference on the sound quality. there are lots of truncation occuring in these algorithms depending on the hardware specifications. slight or significant, but there is...

 

Anyway, I love all of you !

[MorePaul]

Well, you could always do the math to whatever precision you wanted to in high level code, but the management overhead would prob produce a lot of latency -- so I'm not sure that's really very helpful...sorry

[J3RK]

Originally posted by teoman Believe me, I am right ! even simple latency differences in a sound card makes a difference on the sound quality. there are lots of truncation occuring in these algorithms depending on the hardware specifications. slight or significant, but there is... Anyway, I love all of you !

 

Not if the algorithms were specifically written for the platform they are being run on. A decent programmer would most likely take that into consideration. You can't tell me that something like Reaktor 5 on excellent converters does not sound as good as most VAs available.

 

If someone competent is doing the coding, the software will fit the hardware. If the software is written to work well on floating point math, optimized, and well tested, then there shouldn't be a problem. If the software is written for fixed point, and that's what it's run on, then there shouldn't be a problem.

 

Certain types of hardware facilitate creating software for them. Like DSP farms. They make it easy for someone to write a piece of code that works very well. It may be a bit more challenging to write something that's just as good for a general purpose processor, but that doesn't mean that it can't be done. I don't see how you could think otherwise.

[J3RK]

Latency makes zero difference in sound quality. Only in how playable the synth is live.

 

General purpose CPUs have 64 bit integer and double precision 64 bit floating point. A 24 bit Motorola DSP only has a 56 bit accumulator. Of course all of this gets truncated when it reaches the converters anyway. So, the converters are going to be your limiting factor in audio quality. Until it reaches the converters, it's all math. I know I've said that several times, but it doesn't seem to be getting through.

[urbanscallywag]

Make that 32bit integers.