Repro-1 (out now)

[klinik]

So, when vast majority regarded other options more "analog", isn't the only conclusion that the most expensive and well behaving algo isn't very good at sounding "analog"?

No... one could instead conclude that many people are not good at knowing what analog sounds like.

Personally, I think the preponderance of evidence on KVR over the years supports the later conclusion.

You can't conclude anything like that from the results. You'd need to assume there is a generally accepted "analog sound" (against the results of this poll) and that there was a right answer to an opinion poll (against the method of the poll).

If you serve people five variants of a dish and ask "which one of these tastes most french to you", and most people don't choose #3, it's not most french according to the people you asked.

You can pretend that every answer is of equal merit if you want to... But the thing is, there is actual french food. If you asked 20 people who have never had french food and 15 answered that the Pad Thai was most french, the obvious conclusion is that they do not know what french food is. Duh....

My own conclusion from this test is that very few people here knows how true analog synths sound. The majority have a preconceived definition of "analog", "warm", that they asociate with with loss of high frequencies and/or don´t have the equipment/the experience to actually play with real analog synths. Even with my laptop speakers and shitty earbuds I pointed no.3 as the most analog sounding emulation in my first quick test so is not (much) a question of monitors or DAC converters.

[.jon]

You can pretend that every answer is of equal merit if you want to... But the thing is, there is actual french food. If you asked 20 people who have never had french food and 15 answered that the Pad Thai was most french, the obvious conclusion is that they do not know what french food is. Duh....

Tell us what french food tastes like?

[ATN69]

My own conclusion from this test is that very few people here knows how true analog synths sound. The majority have a preconceived definition of "analog", "warm", that they asociate with with loss of high frequencies and/or don´t have the equipment/the experience to actually play with real analog synths. Even with my laptop speakers and shitty earbuds I pointed no.3 as the most analog sounding emulation in my first quick test so is not (much) a question of monitors or DAC converters.

..but still, that is your oppinion. Others may not share it, but hey, thats okay ..we live in a World where people have so many oppinions and Urs asked people what they Think. The result is a reflection of that. It is also a proof that people disagree, at least we can agree on that

[AnX]

You can pretend that every answer is of equal merit if you want to... But the thing is, there is actual french food. If you asked 20 people who have never had french food and 15 answered that the Pad Thai was most french, the obvious conclusion is that they do not know what french food is. Duh....

Tell us what french food tastes like?

Disgusting

[Urs]

Hehe, the original question and poll was never about analogue sound, was it...

There is no such thing as "analogue sound". There's just those filters that *behave* like their analogue counterparts and those that don't. This is why we asked for "analogue behavior", not "analogue sound".

One can use the crappiest synth to recreate some sound from some boutique analogue synth. Takes a day or two to find the right match.

One can use a greatly modeled synth to recreate any sound from some boutique analogue synth. Takes a minute or two.

[EnGee]

You can pretend that every answer is of equal merit if you want to... But the thing is, there is actual french food. If you asked 20 people who have never had french food and 15 answered that the Pad Thai was most french, the obvious conclusion is that they do not know what french food is. Duh....

Tell us what french food tastes like?

Disgusting

Love the Thai food! Oops! Where am I?

[Vesius]

I think there is absolutely a case to say that 5 got such a high vote due to the top end loss/perceived warmth and smoothness as being analogue behaviour.

Sadly, I think this is what the industry has trained people to think by using cheap-ass oversampling filters that rip out huge chunks of the high end that somehow people think it must be better because it uses more CPU ... right?

However, the filter that actually won, 4, has loads of top end. I picked 4 even though it can alias quite badly under some situations specifically because of that top end - although it is true to say I've always favoured grubby and dirty sounds. It's quite possible that I have falsely come to associate the sound of the Pro1 as I think of it with the limitations of how they were typically recorded - specifically I'm thinking mid 90's home studio trance.

In retrospect, it might have been even more interesting to rate these filters in order rather than go for an outright winner. Proportional Representation if you will. I'm pretty certain that 3 might have done a lot better under these circumstances, as a lot of the 1 and 5 vote whilst not liking each other's favourite would have given some vote to 3. Also, it seems the 4 voters put 3 second.

In fact as a bit of fun, we could extrapolate how that might look (if everyone voted in this idealised, completely unrealistic way)

weighting goes (remember, just for a bit of fun):

1st 10 Points
2nd 5 Points
3rd 3 Points
4th 2 Points
5th 1 Point

So, as voting is at the time of posting:

Filter 1 Voters (Many linked 4 and 1 and didn't like 5's lack of character)
Order 14352
1: 680 Points 2: 68 Points 3: 204 Points 4: 340 Points 5: 136 Points

Filter 2 Voters (Guessing here - hard to spot a pattern in 2 voters - but some linked 2 and 5)
Order 25341
1: 21 Points 2: 210 Points 3: 63 Points 4: 42 Points 5: 105 Points

Filter 3 Voters (Typically, not happy with aliasing but want top end)
Order 34521 (5 and 2 in wrong order?)
1: 70 Points 2: 140 Points 3: 700 Points 4: 350 Points 5: 210 Points

Filter 4 Voters (Prioritising character, but not at any expense)
Order 43521 (5 and 2 in wrong order?)
1: 97 Points 2: 194 Points 3: 485 Points 4: 970 Points 5: 291 Points

Filter 5 Voters (Want controlled and smooth sound)
Order 53241
1: 58 Points 2: 174 Points 3: 290 Points 4: 116 Points 5: 580 Points

Final Scoring
1: 926 Points 2: 786 Points 3: 1742 Points 4: 1818 Points 5 :1322 Points
Final Weighting (rounded approx)

1: 14% --- 2: 12% --- 3: 26% --- 4: 28% --- 5: 20%

Now - this obviously won't be right, but I don't think this is a million miles from how things would have been either. The way the original poll was done, does not show that most people who've chosen to speak, and didn't vote 1, would have probably have had 1 as either their fourth of fifth choice. Nor does it show that most people who didn't vote 3 would have quite happily had it as a second choice.

My point is, the original poll doesn't take any account of what most people *don't* want in a filter.

[moscom_electronics]

My point is, the original poll doesn't take any account of what most people *don't* want in a filter.

And that is what matters most!!

[fmr]

You sound like the economists trying to explain an economical crisis they weren't able to predict

[Urs]

We decided to do the poll in public forums because the discussion was more important than the poll results. Had we done a different style of voting (e.g. like Protoverb) we could have gotten more in-depth results - but less discussion. Everything is always about trade offs...

[.jon]

Hehe, the original question and poll was never about analogue sound, was it...

There is no such thing as "analogue sound". There's just those filters that *behave* like their analogue counterparts and those that don't. This is why we asked for "analogue behavior", not "analogue sound".

One can use the crappiest synth to recreate some sound from some boutique analogue synth. Takes a day or two to find the right match.

One can use a greatly modeled synth to recreate any sound from some boutique analogue synth. Takes a minute or two.

To your ears, which filter behaves most analogue

Ears are typically used to receive sound. Actually, that's is the only thing they can do. Hehe.

How does a filter exhibit it's behavior is not by sound? Digital filter can't be measured for power consumption, temperature or other physical attributes, and we don't have access to the internal bitstreams- which leaves us with the obvious conclusion that behaviour == sound in this poll.

[llze]

Ears are typically used to receive sound. Actually, that's is the only thing they can do. Hehe.

they also help you to stay on your feet in balance

[Urs]

How does a filter exhibit it's behavior is not by sound? Digital filter can't be measured for power consumption, temperature or other physical attributes, and we don't have access to the internal bitstreams- which leaves us with the obvious conclusion that behaviour == sound in this poll.

We're entering hair splitting territory

For me sound = "what you can hear" while behaviour = "how the sound changes when settings change"

You need ears for judgement of either, but latter is a more complex matter.

[urosh]

The non-linear function isn't that big of a deal. The bigger deal is solving the linear system imposed by the Jacobian matrix (green area in the PDF)

Ok, so single cycle (and even better, simd) mul and div would make greater impact on overall performance? And if I replace tanh with quite close thing with 2 muls and two adds, but get 8th order system of equations, I might get worse performance than with tanh? (I ask this because I'm pondering should I spend time on investigating particular trick).


BTW, for 4, why did you choose average of two "iterations"/estimates, instead of just taking value of second one? (btw, on first glance, this approach looks to me like fixed point iteration, don't know if it actually is equivalent)

BTW2, have you tried multiple correction steps with 4, for example 4 steps than average? I would be interested in comparison of 32x oversampling vs 4x oversampling + 4 correction steps (should have same processing loads)

[urosh]

double