Diva vs Analogue - a real world test

[Urs]

So you model the overall output of the component so to speak, not the individual factors that lead to that output?

We don't have code that represents an individual component. That's because a component often doesn't have a distinct input and a distinct output. A resistor for example has two equal ends and the current that flows through it can go either way, depending on the voltages present on each side.

What we do have is a set of equations that describe the current over each component based on the voltages around them, and a matrix that connects these in nodes. It's really all about dealing with the relation of voltages and currents in the circuit as a whole - every component can have an effect on any other, hence it has to be looked at as a whole, not as a set of components.

Most filters in DSP literature otoh are based either on the abstract concept of poles and zeros, or they are a set of formulas working on the rather concrete concpet of "sample values" fed into components with distinct imputs and outputs. Former take capacitor impedance into account, latter don't. Former can't model non-linear behavior, but latter can. The method we use (which is similar to the way SPICE works) can do both and is thus closer to the reality of analogue behaviour than either of the standard methods.

You have your development stuff on an online computer?! Well, the good thing is that you always have a backup copy in Utah

Well, we're a dozen people. We need to access our stuff from wherever we are. That said, if anyone gets hold of the source code they'll still have a hard time understanding how it's done

[Z1202]

Earlier in this thread[/url] I posted a couple of "Waldorf Pulse 2" audio clips as examples of the kind of extreme sounds that VA's can't do.

No one spotted that the clips were made using Diva (it's in the track descriptions on soundcloud).

Nice trick. Although I didn't really like Pulse (I had Pulse 1 and sold it years ago). Anyway, your trick motivated me to take another look at Diva and make a short comparison to Monark:
http://vk.com/wall161447783_2957
The comparison didn't have a goal to dial in exactly the same sounds (particularly the 4th example is sounding very different, because the synths do not have identical sets of features and I had to use "similar" rather than "equal" features), rather to compare the abilities of the synths to produce extreme sounds.

The Diva (some 2 years old demo, didn't bother to install the latest one, sorry) was running at the "Great" rather than "Divine" quality, but that was already about the double CPU usage of Monark (the host SR was set to 88kHz).

While it would be interesting to hear comments and discussions, which sounds do you prefer, I do not promise to reveal which synth is which, this is not the purpose of the comparison. Personally, I was now convinced that Diva is also working for the extreme sounds (although I probably will try it a little bit more before possibly convincing myself to get a license), except it was more difficult to use than Monark. Maybe because I'm less familiar with the synth.

[kvaca]

other thing is that analoguesamples /intentionally?/messed with samples in many ways making them harder to compare...

what do you mean? are you referring to me adding noise to Diva examples around a typical analogue noise floor?

not only the analogue noise...you did /but its a question if it was intentionally or by accident/much more:
-all hw samples are typically downsampled which probably more or less affected the whole sound
-some plugin samples use more voices than their hw counterparts which resulted in fuller sound of some Diva examples
-youve used different vibrato for some samples
etc,etc, Im sure to find more discrepancies if I have more time- the most striking one I have desribed in details earlier in this thread/see page 14/ but nobody /including you/ seems interested so far...

honestly, I think all of these discrepancies can potentially have huge impact on peoples preferences...which can make tests like this questionable

anyway thanks for the test

[fluffy_little_something]

So you model the overall output of the component so to speak, not the individual factors that lead to that output?

We don't have code that represents an individual component. That's because a component often doesn't have a distinct input and a distinct output. A resistor for example has two equal ends and the current that flows through it can go either way, depending on the voltages present on each side.

What we do have is a set of equations that describe the current over each component based on the voltages around them, and a matrix that connects these in nodes. It's really all about dealing with the relation of voltages and currents in the circuit as a whole - every component can have an effect on any other, hence it has to be looked at as a whole, not as a set of components.

Most filters in DSP literature otoh are based either on the abstract concept of poles and zeros, or they are a set of formulas working on the rather concrete concpet of "sample values" fed into components with distinct imputs and outputs. Former take capacitor impedance into account, latter don't. Former can't model non-linear behavior, but latter can. The method we use (which is similar to the way SPICE works) can do both and is thus closer to the reality of analogue behaviour than either of the standard methods.

You have your development stuff on an online computer?! Well, the good thing is that you always have a backup copy in Utah

Well, we're a dozen people. We need to access our stuff from wherever we are. That said, if anyone gets hold of the source code they'll still have a hard time understanding how it's done

Ok, so relative component behavior rather than IO. Well, whatever you and a few other companies are doing, it seems to work I find it kind of fascinating that one can simulate electricity so to speak. Simulating air flow around plane wings is much more intuitive to me

(It is OT, but couldn't one also model on a much more abstract level, namely the influence of each and every interface control on the synth's sound (maybe even involving high-quality samples of the sound at control reference points) rather than the behavior of hardware components? )

Maybe the remaining differences that according to the posts here many people still seem to hear are down to equalization rather than emulation.

[Mutant]

Maybe the remaining differences that according to the posts here many people still seem to hear are down to equalization rather than emulation.

Yes. Equalization has a huge impact on how synth sounds are received by human ears/brain.
By coincidence earlier today i did an exercise where i put Synth1, Xhip and Diva on 3 tracks and equalized a very simple saw osc + slightly closed HP filter patch to sound so close that i couldn't tell which track is which synth anymore, while unequalized they sounded quite different.
Of course that doesn't mean that all you have to do is to EQ Synth1 and you will have Diva.
It only works for simple sounds.

Did the same thing half a month ago with Xhip and Jupiter 80.
http://www.kvraudio.com/forum/viewtopic ... 2&t=444026

[SWAN808]

-all hw samples are typically downsampled which probably more or less affected the whole sound
-some plugin samples use more voices than their hw counterparts which resulted in fuller sound of some Diva examples
-youve used different vibrato for some samples
etc,etc, Im sure to find more discrepancies if I have more time- the most striking one I have desribed in details earlier in this thread/see page 14/ but nobody /including you/ seems interested so far...

hw samples downsampled - please explain what you mean here.
plugin samples using more voices - answered below
different vibrato - I can think of one example of this which I doubt would have a significant impact

your comments in page 14. You are referring to an example of polyphonic portamento (not pitch bend). It used the same number of voices. You may hear a difference due to the slightly different portamento rates of the voices...
Your comments about antialiasing filter at the end of that example is incorrect - its a slight difference in the keytracking between synths that was difficult to edit out resulting in a more closed cutoff filter.

[chk071]

Earlier in this thread[/url] I posted a couple of "Waldorf Pulse 2" audio clips as examples of the kind of extreme sounds that VA's can't do.

No one spotted that the clips were made using Diva (it's in the track descriptions on soundcloud).

Nice trick.

No, it isn't. Because that's exactly the sort of thing we don't want in a fair comparison, a biased test, simply made to prove something. What we'd really want is a comparison which really shows and compares the difference there are naturally between analog and digital synths. Won't happen though, because, like with every study or comparison, there'll be bias right from the start, so you can forget the outcome really. But while that is so, the comparison in the OP is still much better than this "trick", which is just BS, just like Synth1 sound demos which should prove something (not sure what though, maybe the superiority of the person who made them?).

[Z1202]

Earlier in this thread[/url] I posted a couple of "Waldorf Pulse 2" audio clips as examples of the kind of extreme sounds that VA's can't do.

No one spotted that the clips were made using Diva (it's in the track descriptions on soundcloud).

Nice trick.

No, it isn't. Because that's exactly the sort of thing we don't want in a fair comparison, a biased test, simply made to prove something. What we'd really want is a comparison which really shows and compares the difference there are naturally between analog and digital synths. Won't happen though, because, like with every study or comparison, there'll be bias right from the start, so you can forget the outcome really. But while that is so, the comparison in the OP is still much better than this "trick", which is just BS, just like Synth1 sound demos which should prove something (not sure what though, maybe the superiority of the person who made them?).

I'd say we want different comparisons. One question is how close the things sound. But that's somewhat academic. The real question is whether the softsynths have sufficient "analog qualities". The AB tests help here, because you might hear a softsynth and think "wow, it sounds cool", but then you hear it's analog prototype and realize that they can't be even compared. So the AB test will help to remove exactly this bias. But whether the things sound indistinguishable or not is beyond the practical point, since even two Minimoogs could sound more different that a model and a real Mini. What we want to compare is not how much the softsynths differ from analog, rather how much of the "analog qualities of sound" do they miss.

[SWAN808]

you dont really need to do trick tests - blind A B tests can accomplish the same thing but without being underhand...when you are dealing with small differences in audio if you bias a test in that way its not really useful data IMO.

I want to say "thank you" to analoguesamples909 for making time to do this test (people usually don't realize that this takes time) - it was fun and VERY usable for me

thanks your kind comments are a welcome tonic

[kvaca]

-all hw samples are typically downsampled which probably more or less affected the whole sound
-some plugin samples use more voices than their hw counterparts which resulted in fuller sound of some Diva examples
-youve used different vibrato for some samples
etc,etc, Im sure to find more discrepancies if I have more time- the most striking one I have desribed in details earlier in this thread/see page 14/ but nobody /including you/ seems interested so far...

hw samples downsampled - please explain what you mean here.

I mean resampling from higher SR down to 44,1kHz using steep low pass filtering which can affect sound in many ways /depending on the quality of processing/

[pepelogu]

A: OB
B: DIVA

If i'm wrong I will buy Diva for sure.

[SWAN808]

I mean resampling from higher SR down to 44,1kHz using steep low pass filtering which can affect sound in many ways /depending on the quality of processing/

run 10 loops thru a decent converter then try a blind ABX test...

[kvaca]

I mean resampling from higher SR down to 44,1kHz using steep low pass filtering which can affect sound in many ways /depending on the quality of processing/

run 10 loops thru a decent converter then try a blind ABX test...

only you know what quality converter was used...thats why I have written "can affect sound"
Im sorry being a little suspicious with this, but while all A examples sounds to me infinitely better/more real/ than B they sometimes have slightly more dull top end than B examples, so I have blamed poor downsampling for that feature...

You are referring to an example of polyphonic portamento (not pitch bend). It used the same number of voices. You may hear a difference due to the slightly different portamento rates of the voices...

if so, can you reveal exact number of voices used?

thank you for all your detailed answers

[lnikj]

^ "infinitely better" ?

[kvaca]

^ "infinitely better" ?

exactly