On Tue, 2012-08-21 at 22:14 +0200, Nick Copeland wrote: Yes, but as already mentioned, for a modular to be usable a standard is required. The analog synths on which this was modeled have such a standard. As it turns out, that was based around A as well, just dug up by Robin Gareus on IRC: https://en.wikipedia.org/wiki/CV/Gate Even volts on A in both schemes. Erm... only being able to modify frequencies upwards would be useless. Nobody is suggesting that *all* "voltages" that affect frequency are absolute. Clearly modulation inputs would be centred about 0 relative to... whatever/nothing. If you do otherwise, then you ruin the ability to modulate frequency with anything (e.g. audio signals). Plugging a e.g. keyboard frequency into an amplifier gain directly (without any conversion) is obviously not going to work no matter how you do things. Using audio to modulate frequency, however, is a fundamental ability. You don't even need modulation inputs for this to be useful, you can just plug the absolute frequency and some audio into the same port (so they will be summed) and get the FM you expect. Regardless, it has to be centered about something. If you want to make it positive only, then it has to be rooted at something. To me it seems all this achieves is limiting the frequency range you can express (since you have to pick a lowest note) and gains nothing. If you divide 440 by 2 successively until you get close-ish to 0, you get numbers like 0.21484375. Centering about standard concert tuning frequency seems quite a bit more sensible to me. If you could start at 0Hz, that would be nice, but of course 0Hz * 2 is... 0Hz :) Fons pointed this out, and while correct in some academic sense, it is not useful. Pretending the absolute frequency signal is not in fact absolute and moving the problem to the receiver doesn't really make the problem go away, and in practice raises a ton of implementation difficulties that weren't there before. Suffice to say it is highly preferable to have *ports* be meaningful on their own. In these modular synths, both analog and digital, signals at times *do* represent absolute frequencies. This is a fact, this is reality, this is not debatable. Yes, you can think of these as all modulation frequencies and move the problem of a base frequency onto the modular itself, which is reasonable in an academic sense, but not useful. It makes the theoretical problem go away by way of semantic trickery only, it doesn't make the actual problem go away. In reality you have some 'note' module which emits CV to describe frequency, which is inherently absolute since describing a frequency is what it does. This means everything needs to agree on what the base frequency is. A good thing for plugins to do would be to have an input that defines the center frequency in Hz, though these plugins do not have such inputs and I'd rather not add ports to them (it is also a teeny bit more math to do, but probably not a significant overhead) Signals inherently have implied semantics at some point if they are ever actually used to do something. No restrictions here. In most apps like this you are always free to plug any signal into anything. The need to define these things is essentially a documentation issue (for both humans and programs). It's all just floats at the end of the day. Thanks for the input, -dr

On Wed, 2012-08-22 at 00:11 +0200, Nick Copeland wrote: It clearly shoes a mapping between volts and absolute frequencies. That there is tuning involved which can modulate the center slightly doesn't matter. The plugins in question also have frequency modulation inputs that can be used for tuning. In the synthesizer as it is working, a voltage is representing an absolute frequency. This is inevitable, because... well, you need to represent an absolute frequency with a voltage when all you have is voltage to convey information. You are correct that it is not set in stone, but it is representing an absolute frequency regardless. Fact: In a real-life genuine analogue Moog, even volts are A, e.g. 4V is A4, or 440Hz +/- tuning. How about a diagram: note.frequency_out ------A------> osc.frequency_in There are 2 possible scenarios here: 1) A carries a representation of an absolute frequency, and both "note" and "osc" agree on what that representation is 2) Your synthesizer does not work It doesn't matter, or even really make sense, to argue about where the semantics of the signal on A "came from". They are there. -dr

On Tue, Aug 21, 2012 at 10:14:28PM +0200, Nick Copeland wrote: I agree. The only way I could maybe understand Dave's view is as follows: You write 'if you set and oscillator to A440'. Now *how* is that done ? It could be some control that is 'internal' to the oscillator, and then all other 'voltages' are relative. Now if that 'internal' control is actually using a port, one *could* say that such a port has an 'absolute' meaning for zero. The weak part in that argumentation is that such a port is not different from any other one that modifies the same frequency. In the end the sum of all values is what matters. Since adding two 'absolute' values makes no sense, there are two ways to look at such a sum: * One term is absolute, the other (N-1) must be relative, or * All N are relative, being added to something absolute which is internal. Since are ports are in fact equal (even if some may have variable gain and some not), I tend to prefer the second way of viewing things. Dave, this is also visible in the code: the only 'absolute' value in the argument of exp2ap() is that constant term, and it doesn't belong to any of the ports whose values are being summed. Ciao, -- FA A world of exhaustive, reliable metadata would be an utopia. It's also a pipe-dream, founded on self-delusion, nerd hubris and hysterically inflated market opportunities. (Cory Doctorow)

On Tue, Aug 21, 2012 at 06:29:22PM -0400, David Robillard wrote: It doesn't have to be true... you can always see the absolute part as a property of whatever receives the signal. Suppose you have a VCO with two 1/octave control ports. One is used by a GUI element, say a slider which could have a scale in Hz, note names, or octaves. The second is the 'voltage' from a keyboard. The actual frequency of the VCO would be F = exp2(V1 + V2 + C) (1) where C is some constant, e.g. log2(440). Each of these two ports can be used without the other, that depends just one what you are doing. In case you use only V1, you could say that C is a property of the port that delivers V1. Same for V2. But if you use both, which one, keyboard or calibrated slider, is the absolute one and which one is the offset ? That, I'd say, is just 'in the eyes of the beholder'. That's why I'd say that both are offsets, and C is a property of the VCO. Ask yourself this: in what way would (1) be different if you'd consider either port to be 'absolute' ? Even if you'd move C 'into a port', the net result would still be the same sum, in which all terms have equal status. Ciao, -- FA A world of exhaustive, reliable metadata would be an utopia. It's also a pipe-dream, founded on self-delusion, nerd hubris and hysterically inflated market opportunities. (Cory Doctorow)

Not sure if I can really comment on these two options but can't you movetuning into the MIDI code? Here is what I mean: > Suppose you have a VCO with two 1/octave control ports.> One is used by a GUI element, say a slider which could> have a scale in Hz, note names, or octaves. The second> is the 'voltage' from a keyboard.> > The actual frequency of the VCO would be> > F = exp2(V1 + V2 + C) (1)> > where C is some constant, e.g. log2(440). The constant 'C' used to be the keyboard voltage from the resistordivider circuit. It was intended to be the voltage that would produceA440 on middle-A on an 8' transpose, no mods or detuning. It was neverthat accurate by which I mean the oscillator signal depended on otherenvironmental influences and is why many of the old beasts had an A440reference sine wave and global/auto tuning. You could extract the same from your MIDI signal so that the MIDI inputis also converted into a port that is summed with the mods to producethe output signal - you are going to have summing, it is needed to linkkbd, mod, fine/rough tuning. Well, in the voltage controlled synths there were properties of anoscillator. If your synth does not have controls but only ports then howdo you select the waveshaping code? How do you select 12+24dB/Oct filters,how do you select keyboard tracking versus non-keyboard tracking oscillators,are these all input ports with implied semantics? Are they different modules? There are other examples but when I was starting on Bristol this single issuelead me to separate out mod/volt controls and property controls since the optionof using an input port to define a constant is a waste of CPU cycles and is alsogoing to be pretty arbitrary. I also liked the concept of it being 'fully bussed' but that truth is that it does not apply itself that well. Having separate modules for each waveform, filter type, tracking methods, etc, is also a waste of time and even more so if different modules have to be invoked each time somebody changes parameters. The other parameter that is going to be an issue is transpose. You couldlook at this as another input port but that is another waste of CPU to havea port for a constant value. Also, the original synths did not have avoltage value that represented transpose, it was a property that adjustedthe rate of the oscillator which drove the waveshaping circuits. The issuehere (with the originals) is that the waveshaping was dependent on thefrequency so if an oscillator is transposed then some changes may alsobe needed to the parameters of the shaping circuit. This is difficult toemulate if you literally only have ports rather than properties/controls.It can be argued here that you are not intending to actually emulate anyspecific synth in which case these interactions are not necessary but ifyou are not emulating then why do you want V/Oct which is an emulation ofthe controls of the analogue synths? Also, if you do want to use V/Oct then perhaps also get rid of -ve values.I mentioned this before and I always get ainsy when considering 'what if thecontrol bus goes negative?' It is unanticipated results for an AMP and -vevalues to a filter cutoff can be, well, put it this way, I check for them.The Moog spec for CV was 0..10v, the value does change according to yoursources but it never included -ve volts. Regards, nick.

On 08/22/2012 01:50 AM, David Robillard wrote: I would prefer 1), because 1/octave might be useful for modulation, but just complicates matters if the task at hand is setting an oscillator to a specific frequency. One could even argue that oscillators should not have 1/octave ports at all, but that there should be a plugin that takes a 1/octave modulation input, a Hz base input and delivers a resulting Hz output. The DRY principal applied to plugins. Even for modulation, a plugin/infrastructure that allows the user to map any value range to a frequency modulation range in semitones could be far friendlier and does not necessarily need 1/octave behind the scenes. I doubt consistency with the original plugins will matter in practice at all. The added workload for you sucks, of course and only you can weight that aspect of it, David. -- Thorsten Wilms thorwil's design for free software: http://thorwil.wordpress.com/

Hi Dave, On Wed, 22 Aug 2012 12:43:27 -0400 David Robillard <d(a)drobilla.net> wrote: Careful there. You will find that to get the 1/oct modulation to work, you'll need a calculation somewhat like this: frequency = exp2(octave_mod_port + log2(Hz_port)) which in light of this overlong discussion, makes me laugh out loud! -Sean

On Wed, Aug 22, 2012 at 12:43:27PM -0400, David Robillard wrote: In fact none of the classical VC synths had anything close to absolute tuning, in the sense that you could accurately set them to some frequency without using a frequency counter, or comparing with another signal. They had to be tuned just like you tune e.g. a string instrument. And most were notoriously unstable as well. It is not necessary in order to have the 'moog' sound - the typical distortion, the way it detunes when overdriven, and the 'noisyness' (which the plugins don't emulate). You need the logarithmic control to emulate the 'look and feel', of a complete analog modular synth, to allow the user to combine modules in similar ways as could be done on the real thing. Ciao, -- FA A world of exhaustive, reliable metadata would be an utopia. It's also a pipe-dream, founded on self-delusion, nerd hubris and hysterically inflated market opportunities. (Cory Doctorow)

On Wed, Aug 22, 2012 at 05:32:26PM -0400, David Robillard wrote: (upper case names refer to the enums in *.h, CV means audio rate in AMS) FREQ (CV, log) : keyboard voltage, so this needs to be log in AMS. OCTN, TUNE (control, log) : from GUI. In fact one control separated into and 'octave' switch and 'tune' slider. They are two ports only because AMS derives the GUI from the plugin interface. Today I'd separate them only in the GUI. Note that there is no problem at all if the for example the VCO has middle C as the reference and the keyboard has A440. Just use the 'TUNE' control to shift everything by 3/4 octave. That's one of the reasons why it exists. Sorry but this is wrong. The filters have: A_FREQ (CV, log) : keyboard voltage), C_FREQ (control, log) : GUI slider A_FMOD (CV, log) : modulation (e.g. envelope), gain set by C_FMODG (slider). So the 'cutoff' is avaiable as both CV and control, and they don't have linear modulation. Ciao, -- FA A world of exhaustive, reliable metadata would be an utopia. It's also a pipe-dream, founded on self-delusion, nerd hubris and hysterically inflated market opportunities. (Cory Doctorow)

On Thu, Aug 23, 2012 at 12:06:06PM -0400, David Robillard wrote: You will probably disagree (so let's agree to disagree :-), but the user of a CV based synth, the real analog thing or an emulation, can be expected to know this. And if not, it's rather obvious what he's supposed to do, once he knows the function of the 'Tune' slider. That sort of thing can be very useful, but it's not what a user of a CV based synth would expect. It's rather what he wouldn't expect. On the real thing, you can connect any source to any destination, TRS sockets are as dumb as they come, and the *only* thing the manufacturer would take care of if that no connection would ever damage the electronics. This is just part of the fun. Even more: on the classic EMS synths which had a pin matrix instead of patch cables, the CV mixing was purely passive. So adding a source to a CV summing bus would actually modify the gain of the existing connections, and you didn't even a have a fixed voltage/octave value. Ciao, -- FA A world of exhaustive, reliable metadata would be an utopia. It's also a pipe-dream, founded on self-delusion, nerd hubris and hysterically inflated market opportunities. (Cory Doctorow)

On Thu, Aug 23, 2012 at 01:07:31PM -0400, David Robillard wrote: Clearly we have different backgrounds. Straying from the technical into the artistic domain is a good way to end this thread, so... All the synth music I really love happens to be made using monophonic instruments. Things like Zawinul's solo 52 seconds into 'A remark you made' <http://www.youtube.com/watch?v=DWvSr2g4U34>. Once synths became polyphonic they mostly turned from a instrument in its own right into something just used to imitate other instruments. I'm working on a SW modular synth that at least in its first version will be strictly monophonic, precisely to emphasize it's meant to be an instrument on it own, and not a cheap replacement for anything else. Ciao, -- FA A world of exhaustive, reliable metadata would be an utopia. It's also a pipe-dream, founded on self-delusion, nerd hubris and hysterically inflated market opportunities. (Cory Doctorow)

On Thu, Aug 23, 2012 at 03:57:50PM -0400, David Robillard wrote: I think that is the best thing to do, and probably should have stated that earlier. BTW, I didn't want to end the thread, but you suggested it was getting too long, so... They are GPL licensed IIRC, so as long as you follow the rules, no objections ! Ciao, -- FA A world of exhaustive, reliable metadata would be an utopia. It's also a pipe-dream, founded on self-delusion, nerd hubris and hysterically inflated market opportunities. (Cory Doctorow)

On Fri, Aug 24, 2012 at 02:32:43PM -0400, David Robillard wrote: In a sense, yes, but the analogy fails in some respects. * It can take years for a player to develop the motoric skills required to e.g. get a decent sound from a violin. But once those are in place, that is something quite difficult to emulate by either analog or digital electronics, yet it is an essential part of any music made on such an instrument. OTOH, an analog synth and a software emulation of the same can have exactly the same human interface, nothing is lost. * There's limit to what can be done with analog eletronics. There's virtually no limit to what can be done in software. A 'perfect' emulation of an analog synth, including all its defects, *is* possible. The brute force method is to build a Spice model of it, and run it on a system that is fast enough to do that in real time. I agree. You need to emulate the function, the visual ergonomics, but that doesn't mean things have to be 'photo-realistic'. Which is why musician prefer hardware controllers. And not only musicians. I find using a 'real' mixer, even a digital one with assignable knobs and switches much more productive than any DAW. Doesn't have to be. The FFT is just a mathematical trick, it doesn't imply anything regarding the sonic results. What makes things sound 'computer' is ignoring the complexity of most real sounds. Ciao, -- FA A world of exhaustive, reliable metadata would be an utopia. It's also a pipe-dream, founded on self-delusion, nerd hubris and hysterically inflated market opportunities. (Cory Doctorow)

[Fons Adriaensen] I wholeheartedly agree with the sentiment; still, some might argue that many ventures that sought to present a replacement for the human voice have left us with a wonderful body of physical instruments (though those of quality certainly rarely are cheap); and others might -- not entirely unreasonably -- claim that trying to emulate the characteristics of one group of instruments using different physical means can yield valuable insights into the science of "what sounds well", regardless of the outcome of the undertaking. Cheers, Tim