SWAY - a synth modeled after the Yamaha SY77

[Monsieur_FyP]

This is ...France !

C'est gentil de nous faire passer pour des attardés incapables saisir ce qu'est un ordinateur

[1darkchild]

Little status update: everything is basically implemented. I'm now in the process of fixing bugs, polishing the GUI, improving performance, and tightening up some measurements, while doing lots of testing. I'm not going to have the final version launched before the end of the year now, as I had hoped. But it shouldn't be too long before I can finally send out a beta version to the testing crew.

How long it takes after that moment depends on how many issues the testers encounter, and on how fast I can make samples to ship with the synth. I know said I would make my own sample set, but I might decide to recreate approximations to the original set after all. Of course it will not sound identical, but what I've managed to create so far sounds close enough in timbre that the AWM patches still make a lot of sense. Going this way would be a huge advantage of course, because there are at least 10,000 SysEx patches to be found online. We'll see how far I can take it.

Count me in for beta testing (FL Studio latest version, Win 11)

[gentleclockdivider]

TO the developer , there is this important section of the tg77 where the awm part can be replaced by the afm part
Which basically means , routing the complete output of the algoritm back to itself ( when afm part is used as second input )


Page 137 of the second part of the manual

[fmr]

I have both the TG77 and the SY99, and I didn't understand what is your point. The synth configuration allows voices (presets) with ONE element (which can be either AFM or AWM), with two elements (which can be either 1 AFM + 1 AWM, or 2 AWM or 2 AFM), and FOUR elements (these are special and can only be stored in Bank D) which can be either 2 AFM + 2 AWM, 4 AFM or 4 AWM. So, you can have as many AWM elements as you want in a voice, depending on the configuration you choose. No need to "replace" anything - in fact what you can always do is replace the configuration (between one of the possibilities listed above). Of course you can always replace an AWM element with an AFM element and vice-versa. You just need to change the voice configuration.

I don't know if you are referring to the fact that you can use the output of the AFM element as a "waveform" in the AWM element. This has the advantage of allowing the AFM element to be processed twice and with two completely different filters, but IMO is not very important (and to do that you will sacrifice the possibility of using a different sound in the second element, which is more important, IMO).

The really important thing, and what makes the SY series a completely different beast way ahead of its time, is that when you have an AWM element, you can route that waveform as a waveform into one of the operators of the AFM element (a process Yamaha called RCM - Realtime Convolution and Modulation).

[gentleclockdivider]

I have both the TG77 and the SY99, and I didn't understand what is your point. The synth configuration allows voices (presets) with ONE element (which can be either AFM or AWM), with two elements (which can be either 1 AFM + 1 AWM, or 2 AWM or 2 AFM), and FOUR elements (these are special and can only be stored in Bank D) which can be either 2 AFM + 2 AWM, 4 AFM or 4 AWM. So, you can have as many AWM elements as you want in a voice, depending on the configuration you choose. No need to "replace" anything - in fact what you can always do is replace the configuration (between one of the possibilities listed above). Of course you can always replace an AWM element with an AFM element and vice-versa. You just need to change the voice configuration.

I don't know if you are referring to the fact that you can use the output of the AFM element as a "waveform" in the AWM element. This has the advantage of allowing the AFM element to be processed twice and with two completely different filters, but IMO is not very important (and to do that you will sacrifice the possibility of using a different sound in the second element, which is more important, IMO).

The really important thing, and what makes the SY series a completely different beast way ahead of its time, is that when you have an AWM element, you can route that waveform as a waveform into one of the operators of the AFM element (a process Yamaha called RCM - Realtime Convolution and Modulation).

Yeah , you don’t understand .
Follow these
Steps , choose and fm+awm part , and choose an algo where the carrier has a free second input .
Go to the operators page and for one of the carriers choose awm at it’s second input and thus modulator
Increase mod depth ( it will,only go to 7 )
The operator is now fm’d by a rom,sample ( which iirc is the saw wave by default )
Now follow the steps,that I have decribed in my previous post ( replace awm by afm)
Result :what do you think is now modulating the operator ?
Exactly , the whole fm structure itself..
I even created a resoant filter this way by routing the structure back to the filter without using resonance ( somehow this method does introduce small amount of latency ..few samples and inverts the signal .)
I,posted it on gearslutz a couple of years ago , have a look , or don’t ..
And there are other small things that might ,not seem so obvious but are welcome to get a decent emulation , and patch compatibility.
For example , did you know that the secondary input of an operator has a lower gain input compared to it’s primary input ?
Another one , just like the microwave , the tg77 has some kind of asic bug where an operator can clip and fold backs to,itself when exceeding 127;velo value

[IvyBirds]

The operator is now fm’d by a rom,sample ( which iirc is the saw wave by default )
Now follow the steps,that I have decribed in my previous post ( replace awm by afm)
Result :what do you think is now modulating the operator ?
Exactly , the whole fm structure itself..

In the real world I have never found doing that to be all that practical. Have you? It's a fun exercise to play around with but I haven't found it all that useful, especially in 2024

With FM-X inside of Montage you can do similar things even use the external audio in as a modulation source, and have far more options at the operator level for waveshaping, feedback, not to mention having 8 operators to use

In my experience the samples in SY77/TG77 have to many harmonics and are to complex to be a good modulation source. Having the ability to load in your own samples would be useful for that, but then I would end right back up with more simple waveforms that I can just use a couple of operators to create anyway

I spent a while using the Timbre Frames features from the Synclavier II to make evolving simple waveforms using the additive/FM engine and then feeding that into the audio in of Montage as a modulation source that was yielding interesting results

In the end it was just to much to tame and I abandoned the idea as I really wasn't able to accomplish anything I couldn't get to from other methods

With a SY77 plugin it might be easier to accomplish than it was on my TG77 hardware, so I will revisit the idea

[sheaf]

Now follow the steps,that I have decribed in my previous post ( replace awm by afm)
Result :what do you think is now modulating the operator ?
Exactly , the whole fm structure itself..

I have included the AWM into AFM and AFM into AWM options of course. But if you create a feedback loop by using both at the same time it only seems to accurately reproduce what the hardware does up until input level 5 (with all other relevant levels set to maximum). At input level 6 and 7 my TG77 turns it into a noisy mess with some resonant peaks.

This latency you mentioned is always there between elements in any configuration. Create a normal 2AWM patch with a sawtooth wave where both elements are identical. You'll see and hear comb filtering going on caused by the latency. What's worse, this latency is not identical for every note on event. I'd say it's about 60 samples 80% of the time, but on some notes it's more and on some it's less. And there is about 120 samples between element 1 and element 3, and even more between 1 and 4 (not 180 though as you might expect, that would make too much sense).

For example , did you know that the secondary input of an operator has a lower gain input compared to it’s primary input ?

You mentioned this before, but I wasn't able to reproduce it. And the manual says it's not the case. Any way you can show this?

Edit: Found your gearspace thread on how to reproduce it. I'll investigate.

[sheaf]

I think I know what's going on, and it's not about a difference in input gain 1 vs 2. When you select a feedback input that is fed by an operator higher up in the stack, you might expect it's really a feedforward path, and that seems to be your assumption too. But it does actually select the feedback path. You can use the feedforward path too, but only using SysEx. In SWAY you can easily do freeform editing though.

But why does the feedforward version even sound different from the feedback version, isn't it just a 1 sample delay? No, because the feedback paths are filtered by a filter with transfer function H(z) = 1/2 + 1/4z. This reduces the volume and so you get less modulation.

[gentleclockdivider]

Now follow the steps,that I have decribed in my previous post ( replace awm by afm)
Result :what do you think is now modulating the operator ?
Exactly , the whole fm structure itself..

I have included the AWM into AFM and AFM into AWM options of course. But if you create a feedback loop by using both at the same time it only seems to accurately reproduce what the hardware does up until input level 5 (with all other relevant levels set to maximum). At input level 6 and 7 my TG77 turns it into a noisy mess with some resonant peaks.

This latency you mentioned is always there between elements in any configuration. Create a normal 2AWM patch with a sawtooth wave where both elements are identical. You'll see and hear comb filtering going on caused by the latency. What's worse, this latency is not identical for every note on event. I'd say it's about 60 samples 80% of the time, but on some notes it's more and on some it's less. And there is about 120 samples between element 1 and element 3, and even more between 1 and 4 (not 180 though as you might expect, that would make too much sense).

For example , did you know that the secondary input of an operator has a lower gain input compared to it’s primary input ?

You mentioned this before, but I wasn't able to reproduce it. And the manual says it's not the case. Any way you can show this?

Edit: Found your gearspace thread on how to reproduce it. I'll investigate.

Take algo 34 , a simple two tower structure with 3 operators each



Let's ignore operators 2 and 3 , operator 6 is also disabled in this example
NOw , if we use the feedback (in this case it's feedforward ) routing to route the output of operator 5 to the second input of operator 1(input level set to +7),




Now operator 1 should produce the exact same output as operator 4 .
This is sadly not the case , somehow the secondary input of an operator does some kind of gain reduction at it's input , even when set to max +7.
Here's two audio files
First is right structure only , so only operators 4,5 , second file is modified structure , operator 1 takes op 5 at it's second input (op2,3,4 muted )
Attached Files
https://gearspace.com/board/attachments ... ty-one.wav

https://gearspace.com/board/attachments ... ty-two.wav

Link to gearlutz post and confirmation by other users
https://gearspace.com/board/electronic- ... ocity.html

[gentleclockdivider]

I think I know what's going on, and it's not about a difference in input gain 1 vs 2. When you select a feedback input that is fed by an operator higher up in the stack, you might expect it's really a feedforward path, and that seems to be your assumption too. But it does actually select the feedback path. You can use the feedforward path too, but only using SysEx. In SWAY you can easily do freeform editing though.

But why does the feedforward version even sound different from the feedback version, isn't it just a 1 sample delay? No, because the feedback paths are filtered by a filter with transfer function H(z) = 1/2 + 1/4z. This reduces the volume and so you get less modulation.

INteresting, so the feedforward routing are actually feedback ?
I don't undestand what you mean by only sing sysex .
This is feedforward , and can be done directly from the panel

[sheaf]

I think I know what's going on, and it's not about a difference in input gain 1 vs 2. When you select a feedback input that is fed by an operator higher up in the stack, you might expect it's really a feedforward path, and that seems to be your assumption too. But it does actually select the feedback path. You can use the feedforward path too, but only using SysEx. In SWAY you can easily do freeform editing though.

But why does the feedforward version even sound different from the feedback version, isn't it just a 1 sample delay? No, because the feedback paths are filtered by a filter with transfer function H(z) = 1/2 + 1/4z. This reduces the volume and so you get less modulation.

INteresting, so the feedforward routing are actually feedback ?
I don't undestand what you mean by only sing sysex .
This is feedforward , and can be done directly from the panel

If you do it by sending feedback path 2 (sourced from op5) using the panel it says "FBOp5" at "In2 Src". This is feedback mode, not feedforward. Is this what you see too? Internally in the hardware feedback path 2 is represented by the number 7.

If you want to use feedforward mode you need to set "In2 Src" to feedforward path 2, which is internally represented by the number 4. If you do this it will show "OP5" at "In2 Src". You can't do this from the panel as far as I know (unless I'm missing something?). But you can do it by sending a patch through SysEx where you utilize the number 4 instead of 7.

[sheaf]

fb_vs_ff.jpg

I realized what I wrote might be too confusing so maybe it's better if I show it.

The left side shows OP5 going into OP1 using feedback mode. I assume this is what you're doing, while thinking it's feedforward mode.

The right side shows OP5 going into OP1 using actual feedforward mode, which can't be done from the panel (I think, someone correct me if I'm wrong).

In feedback mode you get a less strong FM modulation because of the feedback filtering. In feedforward mode it sounds exactly the same as the configuration enabling OP4 instead of OP1.

[gentleclockdivider]

The confusion is just that the free routing are always named fb , that’s just the nomenclature that is used .
If there is a free input available and an operator higher up the stack is routed in that input it is feedforward .
If the output of the receiver is routed back again , then we have feedback .

[sheaf]

You're right that technically there is no feedback going on either way. But when the TG77 panel reports it's using feedback as an input you can be sure it's not a straight input from an operator higher up the stack. That's what I'm trying to explain. To do that you need the other input source types that are not available on the panel. When you use those you will notice the display says "OP5" instead of "FBOp5". And in that case the modulation will be just as strong as when using a previous operator input.

[cslevine]

( i think about something : maybe FM modulation using samples as modulator does need some little latency, so that the carrier "needs to know" the shape of the wave a little ahead, to be modulated ? )