With All These Emulations Coming Out...

[pdxindy]

And any A6 sample library is fundamentally recordings of the synth at particular settings. Once sampled, much of the interaction that makes the instrument interesting has already been frozen. You can animate the sample afterwards, but you are animating a photograph of the system rather than operating the system itself.

Exactly... Samples entirely miss the lovely hands on experience of tweaking a sound to taste on the fly. They are static snapshots of a dynamic system.

[IvyBirds]

The A6’s variation emerges throughout the entire voice and signal path, including places the programmer may not have explicitly modulated.

In April of 2001 SOS magazine said the following and they were spot on

"Despite Alesis' claims to the contrary, the Andromeda 16-Voice Real Analogue Synthesizer (as they call it) is not a real analogue synthesizer — at least, not completely. Huge chunks of its architecture are digital, and I'm not only referring to the digital effects, the microprocessor-controlled operating system, or even the memories. No... the controllers in the sound-generation system — the envelopes, the LFOs, and the Sample & Hold — are all digital, as is the oscillator tuning (although not the oscillators themselves). This makes the Andromeda a hybrid analogue/digital synth"

https://www.soundonsound.com/reviews/al ... -andromeda

If you take a look at the A6 service module you will see it has Motorola MCF5307 32-bit ColdFire microprocessor running at 90MHz as it's brain. That means when you talk about the envelopes they are 100% generated with a computer as digital data, when you talk about pitch you are 100% talking about digital data, every time you press a key the onboard computer reads that and generates digital data to say hey we need to play a waveform at 261.63 Hz because they pressed the C Key. Every point on every envelope is a piece of digital information, every LFO and modulation source is 100% generated as digital data and some of those are digital data manipulating other digital data in the CPU

From there the digital data is converted with Burr-Brown PCM54HP DAC chips.These are highly precise, 16-bit audio-grade DACs. These are not introducing any errors or variability

From the DAC chip the freshly created analog voltage gets routed to a capacitor where it's stored. The CPU cycles back around and re-charges ("refreshes") the capacitor with a updated voltage value. From there it goes to an Op-Amp which smooths everything out and from there to the VCA or whatever

Because the DAC chip must sequentially update parameter after parameter, the modulation signals sent to the analog chips are technically infinitesimally out of phase with one another. While they happen within microseconds, they do not hit the analog components at the exact same physical native time.

All of that means it's a 100% digital envelope that is being modulated by nature of how the DAC works alongside the capacitors and the op-amps so by shifting your sample playback times by microseconds per key press in a perceivably random way you also inherently make everything infinitesimally out of phase with one another with the exact same audible result

Beyond that of course is the effect that temperature can have here which is why with Andromeda, Alesis gave it an auto tune or auto calibration feature to deal with drift created by fluctuating temperatures. In a perfect world where the temperature would be 100% stable all the time you wouldn't need that, but essentially tempeture is acting here as a modulation source that can effect the frequency of waveforms coming out of the oscialtors. That's all it can do. If you press A keys it should output 110, 220, 440 Hz , etc but they drift and instead output perhaps 108, 222, and 439 Hz. Since tempeture is modulating that in a random way any random modulator can also do so and as long as it is doing so with same result regarding frequency it will have the same exact result.

And any A6 sample library is fundamentally recordings of the synth at particular settings. Once sampled, much of the interaction that makes the instrument interesting has already been frozen. You can animate the sample afterwards, but you are animating a photograph of the system rather than operating the system itself.

If you are talking about finished samples patches sure, but I am specifically talking about raw samples of the oscialtors. On Andromeda make the filter be wide open, turn off resonance, turn off the filter envelope, and on the VCA set sustain to 100, and everything else to 0. Set a single Oscillator to whatever you want and turn off the second one. Play it back and use a Ladder filter and/or SEM filter in Falcon or Omnisphere and apply the apply the Modulation I spoke of earlier and you will get remarkably close to an Andromeda. It won't sound exactly the same because Falcon and Omnisphere both do a significantly better job at modelling a Moog Ladder Filter, and an Oberheim SEM type filter than Andromeda does and they are not specifically trying to mimic the suspect job that Alesis did back in the day

Saturation can also be dialed in if that is an effect you are going for in a myriad of ways

[kraster]

The A6’s variation emerges throughout the entire voice and signal path, including places the programmer may not have explicitly modulated.

In April of 2001 SOS magazine said the following and they were spot on

"Despite Alesis' claims to the contrary, the Andromeda 16-Voice Real Analogue Synthesizer (as they call it) is not a real analogue synthesizer — at least, not completely. Huge chunks of its architecture are digital, and I'm not only referring to the digital effects, the microprocessor-controlled operating system, or even the memories. No... the controllers in the sound-generation system — the envelopes, the LFOs, and the Sample & Hold — are all digital, as is the oscillator tuning (although not the oscillators themselves). This makes the Andromeda a hybrid analogue/digital synth"

https://www.soundonsound.com/reviews/al ... -andromeda



If you take a look at the A6 service module you will see it has Motorola MCF5307 32-bit ColdFire microprocessor running at 90MHz as it's brain. That means when you talk about the envelopes they are 100% generated with a computer as digital data, when you talk about pitch you are 100% talking about digital data, every time you press a key the onboard computer reads that and generates digital data to say hey we need to play a waveform at 261.63 Hz because they pressed the C Key. Every point on every envelope is a piece of digital information, every LFO and modulation source is 100% generated as digital data and some of those are digital data manipulating other digital data in the CPU

From there the digital data is converted with Burr-Brown PCM54HP DAC chips.These are highly precise, 16-bit audio-grade DACs. These are not introducing any errors or variability

From the DAC chip the freshly created analog voltage gets routed to a capacitor where it's stored. The CPU cycles back around and re-charges ("refreshes") the capacitor with a updated voltage value. From there it goes to an Op-Amp which smooths everything out and from there to the VCA or whatever

Because the DAC chip must sequentially update parameter after parameter, the modulation signals sent to the analog chips are technically infinitesimally out of phase with one another. While they happen within microseconds, they do not hit the analog components at the exact same physical native time.

All of that means it's a 100% digital envelope that is being modulated by nature of how the DAC works alongside the capacitors and the op-amps so by shifting your sample playback times by microseconds per key press in a perceivably random way you also inherently make everything infinitesimally out of phase with one another with the exact same audible result

Beyond that of course is the effect that temperature can have here which is why with Andromeda, Alesis gave it an auto tune or auto calibration feature to deal with drift created by fluctuating temperatures. In a perfect world where the temperature would be 100% stable all the time you wouldn't need that, but essentially tempeture is acting here as a modulation source that can effect the frequency of waveforms coming out of the oscialtors. That's all it can do. If you press A keys it should output 110, 220, 440 Hz , etc but they drift and instead output perhaps 108, 222, and 439 Hz. Since tempeture is modulating that in a random way any random modulator can also do so and as long as it is doing so with same result regarding frequency it will have the same exact result.

And any A6 sample library is fundamentally recordings of the synth at particular settings. Once sampled, much of the interaction that makes the instrument interesting has already been frozen. You can animate the sample afterwards, but you are animating a photograph of the system rather than operating the system itself.

If you are talking about finished samples patches sure, but I am specifically talking about raw samples of the oscialtors. On Andromeda make the filter be wide open, turn off resonance, turn off the filter envelope, and on the VCA set sustain to 100, and everything else to 0. Set a single Oscillator to whatever you want and turn off the second one. Play it back and use a Ladder filter and/or SEM filter in Falcon or Omnisphere and apply the apply the Modulation I spoke of earlier and you will get remarkably close to an Andromeda. It won't sound exactly the same because Falcon and Omnisphere both do a significantly better job at modelling a Moog Ladder Filter, and an Oberheim SEM type filter than Andromeda does and they are not specifically trying to mimic the suspect job that Alesis did back in the day

Saturation can also be dialed in if that is an effect you are going for in a myriad of ways

Sure, the control signals are calculated digitally, but they are operating on analogue VCOs, filters and VCAs.

The control signal is the target; the response of the analogue circuit is the result.

An oven where you control the heat digitally doesn't make the resulting heat "digital"

It is this response that is meaningful, not how the control signal originated. That is what makes it analogue.

Would you say that a Minimoog's pitch being controlled by a MIDI-to-CV converter is a digital synth? No. It is still a Minimoog being controlled by control voltage, even if the provenance of that control signal is ultimately digital. You still get all the attendant behaviour of an analogue synth because the VCOs themselves are analogue.

That is more or less what the Andromeda is doing internally. It has never been a secret.

A sample, by definition, is a fixed recording. Left unmodified, it reproduces the same captured waveform each time. An analogue voice, by contrast, regenerates the sound through physical circuitry each time it is played.

Component tolerances alone mean that the voices in an analogue polysynth will not respond identically. The Andromeda uses eight oscillator ASICs and eight filter ASICs, with each chip serving two voices, and each physical voice will exhibit slight differences in behaviour. This is true of analogue polys generally, and it is the principle characteristic people are referring to when they talk about voice variation in analogue polysynths.

Temperature can affect the sound in all kinds of ways: oscillator scaling, waveform symmetry, filter cutoff, resonance, duty cycle, gain, bias and clipping behaviour, often differently across voices.

All of this cannot be reduced to a random LFO applied to oscillator pitch. It is complex, system-wide behaviour.

The Andromeda has two continuous tuning modes. One runs in the background on voices that are not currently being played, and the other applies temperature compensation. There is also a more comprehensive Auto Tune routine that calibrates the VCOs, pulse width, filters and other analogue circuitry. All of these behaviours are defeatable by the user.

A sample will always be the same. Every time since it's a specific snapshot

While Omnisphere, Falcon and others may model analogue filters very well, and you may prefer them, the undeniable fact is that the Andromeda is not “modelling” its filters at all. They are actual analogue filter circuits inspired by SEM and Moog ladder topologies. Again, whether you like them or not is completely subjective but they are not digital representations of analogue circuitry, they ARE analogue circuitry.

Your point about saturation misses the point entirely. Saturation in an analog synth happens at multiple stages, summing of the waveforms of each oscillator, pre filter level driving the filter, filter feedback, post filter level, the post filter distortion circuit and the output of the synth itself. Those nonlinearities are cumulative and state-dependent. The behaviour of each stage depends on what is feeding it and how hard the preceding stages are being driven and each stage will also have minute variations across multiple voices.

In principle DSP can model all of this, but doing it comprehensively would require detailed nonlinear models of every stage, running at sufficient internal oversampling, with separate component and state variation for every voice. That is an enormous real-time computational problem and well beyond what most software synths or sample-based instruments actually attempt.

None of this is required in the Andromeda because the signal path is analogue.

Slapping a generic distortion algo on a sample after the fact comes nowhere near this.

[IvyBirds]

Sure, the control signals are calculated digitally, but they are operating on analogue VCOs, filters and VCAs.

Sure and I explained that. I also explained why that doesn't mean as much WITH ANDROMEDA as you think it does

The control signal is the target; the response of the analogue circuit is the result.

Sure and the computer is constantly monitoring those and correcting for errors

An oven where you control the heat digitally doesn't make the resulting heat "digital"

With an oven the end result is heat, with synths that are digital and analog the end result is an analog voltage that will be amplified and cause speakers to vibrate which causes air to vibrate so we can hear the results

If you use a digital thermostat in an oven it can only do a few things. It can turn the burner or coil on and off or it can increase it decrease the temperature output

With synths you always have some kind of "Oscillator" that can be a VCO, a DCO, a Sample, a Wavetable, etc and all you can do is turn it on and off, or change the frequency of the output. That's it. If a computer is doing that control it will be as precise or as loose as the programming. If those controls are 100% analog which they are not on Andromeda the envelope generators are not very precise and can go out of calibration as can the thing they are controlling so you have the opportunity for double errors

It is this response that is meaningful, not how the control signal originated. That is what makes it analogue.

Would you say that a Minimoog's pitch being controlled by a MIDI-to-CV converter is a digital synth?

The issue with this analogy which blows up your narrative regarding Andromeda is that the Minimoog in your example doesn't have digitally generated envelopes or digitally generated modulators, you are taking incoming MIDI note data and turning into a voltage which means the note data would be every bit as digital as the Andromeda is for the frequency response if the oscialtors

Andromeda has digital everything which makes it a digital synth. When you press a key on an Andromeda that registers with the CPU which then generates digitally everything, every control needed for the patch, the frequencies of the oscialtors, the point values on the envelopes, the modulators etc. That doesn't happen with your vintage Minimoog example

It is still a Minimoog being controlled by control voltage, even if the provenance of that control signal is ultimately digital. You still get all the attendant behaviour of an analogue synth because the VCOs themselves are analogue.

Your skipping over everything else that makes a vintage Minimoog an analog synth that is radically different from Andromeda, and you are skipping over the fact that Andromeda has auto calibration and correction features to keep the system inline and behaving

That is more or less what the Andromeda is doing internally. It has never been a secret.

No it's not a vintage Minimoog and Andromeda are RADICALLY different internally.

A sample, by definition, is a fixed recording. Left unmodified, it reproduces the same captured waveform each time. An analogue voice, by contrast, regenerates the sound through physical circuitry each time it is played.

Sure and if I make a recording of those waves going through the physical circuitry and play it back it will cause speakers to vibrate in the exact same way as the synth which means air will vibrate in the exact same way, which means humans will hear it in the exact same way.

Component tolerances alone mean that the voices in an analogue polysynth will not respond identically. The Andromeda uses eight oscillator ASICs and eight filter ASICs, with each chip serving two voices, and each physical voice will exhibit slight differences in behaviour.

Now we are getting somewhere. You just got to the heart of the matter. With Andromeda everything is generated digitally and then gets sent to the various components. Those components will have a certain tolerance as to how accurate they respond to the incoming data and the computer inside of Andromeda corrects itself to hit those tolerances

So if the computer wants a value of say 14 it will send a value of 14 to that component. That component might have a tolerance where 13 or 15 is also acceptable. Meaning the value of that component will always be between 13 and 15. That drift between 13 and 15 can be accomplished entirely in the digital domain as well. It's super easy for modern software to dial in those variances as we discussed before.

If the component is defective or out of calibration it could be further out of tolerance, in that case the onboard computer if Andromeda will adjust it's control signal to try to make it come back to spec. This is also behavior rather easy to emulate in a modern plugin and of course with a modern plugin you can dial in the exact range you want these things to drift from the norm

This is true of analogue polys generally, and it is the principle characteristic people are referring to when they talk about voice variation in analogue polysynths.

Sure but voice variation is also extremely easy to mimic in software, you are acting like it's some kind of magic unicorn dust. It's not. They are just standard deviations away from a norm. If they exist they can be measured and if they can be measured they can be emulated some plugins do a very good job of this. Modern hardware polysynths even have things like vintage knobs that use digital controls in the exact same way to introduce this

Temperature can affect the sound in all kinds of ways: oscillator scaling, waveform symmetry, filter cutoff, resonance, duty cycle, gain, bias and clipping behaviour, often differently across voices.

Sure and all of those things can be measured and emulated. Many of them are also extremely subtle to the point of being imperceivable

And again if you can hear those things you can record those things and if you can record those things you can sample those things and hear it when you play back your sample

All of this cannot be reduced to a random LFO applied to oscillator pitch. It is complex, system-wide behaviour.

Never ever did I say it could, I have always advocated for a complex system wide behavior. Where you are doing a lot more than introduce a single LFO applied to pitch.

The Andromeda has two continuous tuning modes. One runs in the background on voices that are not currently being played, and the other applies temperature compensation. There is also a more comprehensive Auto Tune routine that calibrates the VCOs, pulse width, filters and other analogue circuitry. All of these behaviours are defeatable by the user.

And all of those behaviors can be emulated easily in software. Why? Because they are computer controlled software from the start. If you turn them off awesome now you have a a bunch of components that might be further out of spec, again if you can hear it, you can measure it and if you can measure how far they are out if whack you dial that in

A sample will always be the same. Every time since it's a specific snapshot

Only it's not as has been explained before. With Andromeda you could also say that a Sawtooth wave is also always a Sawtooth wave and will always be the same. Unless it's modulated by something else. If I randomly modulate the phase of the sample it's different every time, if I alter the playback speed of the sample by micro amounts it's different every time

While Omnisphere, Falcon and others may model analogue filters very well, and you may prefer them, the undeniable fact is that the Andromeda is not “modelling” its filters at all. They are actual analogue filter circuits inspired by SEM and Moog ladder topologies.

And? What you are saying is that Alesis modeled the ladder filter from moog with digitally controlled analog circuits and the SEM filter with digitally controlled analog circuits and did a poorer job of their model than UVI or Spectrasonics did

Again, whether you like them or not is completely subjective but they are not digital representations of analogue circuitry, they ARE analogue circuitry.

Sure analog circuitry 100% controlled and calibrated by digital signals only to achieve a poorer result

Your point about saturation misses the point entirely. Saturation in an analog synth happens at multiple stages, summing of the waveforms of each oscillator, pre filter level driving the filter, filter feedback, post filter level, the post filter distortion circuit and the output of the synth itself. Those nonlinearities are cumulative and state-dependent. The behaviour of each stage depends on what is feeding it and how hard the preceding stages are being driven and each stage will also have minute variations across multiple voices.

I would say you are missing the point. If your goal is saturation and color you can get that in any myriad of ways, as to which is more pleasant to your ears that is entirely subjective much like which model of the Moog Ladder Filter is better the one in Omnisphere or the one in Andromeda

In principle DSP can model all of this, but doing it comprehensively would require detailed nonlinear models of every stage, running at sufficient internal oversampling, with separate component and state variation for every voice.

Who said anything about adding a generic distortion algorithm on a sample after the fact? That's the overall issue here, you are using your confirmation bias to dumb down every possible scenario to fit your preconceived notion that it can't be done and by default Andromeda will always be better no matter what

There will always be differences between a decades old polysynth and a virtual instrument no one said there would not be, most certainly not me

All I am saying is that pretty much everything the hardware crowd seems to think is so magical can easily be done with modern samplers. There really isn't any magic unicorn fairy dust, this is especially true with Andromeda because at its heart it's a digital workstation where a computer is generating everything

[kraster]

I think you’re doing a fair bit of category-juggling here.

Yes, the Andromeda is digitally controlled. Nobody has disputed that. But you keep turning “the CPU generates control values” into “the CPU generates everything,” which is a pretty heroic leap.

It doesn’t generate the audio waveform. The VCOs do that. It doesn’t calculate the filter response. The analogue filters do that. It doesn’t simulate the VCAs. The VCAs are actually there.

That distinction is not exactly a minor technicality.

The speaker argument is a complete dead end as well. Spotify also ends up as an analogue voltage moving a speaker cone. By that logic, Spotify is an analogue synth and we can all go home.

The calibration system is getting promoted into something rather more impressive than it actually is too. It corrects specific things like tuning, pulse width and filter tracking. It is not constantly patrolling all sixteen voices, ironing out every difference in waveform shape, resonance, gain, bias, clipping and saturation like a tiny digital janitor.

And the “target 14, actual 13–15” example makes analogue circuitry sound like a spreadsheet with an RNG attached.

The response changes with level, frequency, temperature, circuit state and what the previous stage is doing. The stages interact. It is dynamic behaviour, not just “add 3% vintage wobble.”

It certainly can be modelled in principle. Nobody is invoking magic unicorn dust. But you keep sliding from “it can be measured” to “it is super easy to emulate” as though those are the same claim. The saturation simulation alone at this level would take some fair CPU grunt as explained before. Every interaction along the signal chain with oversampling and you would already be eating into a significant portion of your CPU budget.

The sampling argument has the same problem. A sample captures one result. Changing its pitch, phase or playback speed gives you altered versions of that recording. It does not suddenly turn the recording back into the system that produced it.

It will not recreate the way oscillator level changes filter drive, resonance changes headroom, or several nonlinear stages push and pull against one another. It is still a recording being manipulated.

And again saying Alesis “modelled” the Moog and SEM filters in the same sense as Omnisphere is doing some fairly impressive acrobatic work with the word “modelled.” They built physical analogue circuits inspired by those designs. Omnisphere runs DSP intended to imitate them. Those are not the same thing just because the same verb has been shoehorned into both.

You may prefer software. You may think Omnisphere sounds better. Fine. That is completely subjective.

At this point your argument is basically: there’s a computer inside it, therefore the analogue circuitry doesn’t count. Which is silly.

[El°HYM]

There even is a 12 / 14 bit DAC inside the Jupiter 8, also a CPU (Zilog Z80). I guess Ivy just loves typing a bit more than reading.