Instruments • Re: Dawesome MYTH

"I know Vital also has an external free resynthesis-- or maybe more accurately called inference?-- plugin too." ~ Borbolactic

"I've seen this, it is more akin to what synplant is doing, in that it uses the engine its self, rather than an array of oscillators. However, from everything I've seen, it is incredibly mediocre, I am sorry to the creator of this tool, but it seems to do poorly at replicating even the most basic sounds. Stuff that you shouldn't even need resynthesis for to begin with." ~ Databroth

What is an 'engine' and since when are the oscillators not part of it?

Also, AFAIK, AI is less about 1-to-1 replication and about inference or interpretation, rather like if you're an AI and I play you a song and you whistle it back to me. Not a perfect replication, but I get something different yet similar to work with.

"If my interest/aesthetic is along the lines of accurate replication or better (think hyperrealistic painting perhaps, and upping the colours and contrast) and a deep, maybe per-partial, control of the sound from there, what might you recommend?" ~ Borbolactic

"Neither Myth, nor Synplant give you 'per-partial control over the sound'.
It's hard for me to make suggestions because the sound material you are using is just as important as your goals. None of the resynthesis tools out there are perfect, they will all stray from various types of sounds, and vary in the complexity of control you can achieve from there." ~ Databroth

What's the best that gives you a good level of per-partial control with resynthesis and/or AI 'resynthesis'/'inference' do you think? Harmor? Is that still the best? Seems odd if it is, yes? Over 10 years old? Virsyn's Cube 2? Parsec 2 or Razor with those stupid rack/reaktor matryoshkas? Neither of the latter 2 seem to do resynthesis. And Harmor's on the way out WRT it being free from being matryoshkafied, such as within a DAW.

Someone here says HALion7. But, 'sample-based'? What does that mean? Doesn't look like it does additive/per-partial though. Pigments? Spectral engine any good? Does it do resynthesis? How pricey is UVI Falcon?

"before you read this next paragraph, please know that I am no expert on AI:" ~ Databroth

I somehow doubt you'd be on here if you were. 'u^ <-- right-side-up winkie

"As for the resynthesis and AI, they are all different, often 'AI' in this context means machine learning." ~ Databroth

Or 'artificial intelligence', coupled or including, perhaps, LLM's (large language models) and whatnot. I guess music/sound models/training come into play somewhere in there.

"There is often a model that is trained off of many sound sources, this training is developed over many cycles using a data-set and some sort of sound engine. In synplant, this sound engine is the synth of synplant, so when training, it is using synplant, adjusting controls, and trying to match the sound source. After enough training this is included with the synth as the 'AI model'. The model has some idea of what being 'close to the sound source' is supposed to sound like, but no idea what your particular soundsource will be.
So you import a sample and it goes through a process of trying to get close to that sound source. By the end of this process it gives you a result, the accuracy of this result depends on a few factors, how closely it matches any of the training data, if the engine could even replicate the sound in the first place, how complex the sound is. It is not FFT at all, though 'fft' may play a role in the analysis of the generation to determine if the results are similar or not." ~ Databroth

Sure. Likely a bit like RVC (realtime voice cloning?). I had it on my laptop before Bit Locker locked me out of my own computer. I'm still recovering and in therapy.

"With Myth, it is very similar in terms of the 'AI' training side of things, but the difference is the engine, and the layer splitting. Myths engine (as far as the resynthesis Iris) is based on a complex bin array of many parallel oscillators, this is similar, but not quite 'fft'. The training and generation are similar, but you have a very different engine involved, so different sources have better potential to be replicated. With Myth, the idea isn't perfect replication anyways, 'it's not meant to be a sampler' as Peter has said. It is meant to give you a creative starting point that is unique from traditional oscillators." ~ Databroth

Any link to Myth's manual online that I could take a read? Is Myth ultimately chopping up the inference/resynthesis into grains, perhaps like its other synths or are we getting true synthesis after the fact of the resynthesis? I ask this because granular 'synthesis' doesn't quite seem like true synthesis and more like sampling at the granular level. What do you think? 'Bin arrays'? What's in them? How do they work?

"I think it's too easy to get caught up in resynthesis and AI, it's a LOT of work to go through to not do any sound design, one of the most fun parts of using synths. It's already incredibly easy to not do any sound design, just use presets, or just use samples." ~ Databroth

There are many ways to skin a cat. You can do sound design before the resynthesis, perhaps using layered pots, pans and glasses in the kitchen (maybe with a baby crying for special IR for convolution reverb), or after its been resynthesized. Your call, yes? Fun to be had by all, no mater how approached, yes?

"Perhaps this is because Synplant uses FM oscillators? (I wonder what kind of oscillators Myth uses.)" ~ Borbolactic

"well there's not really such a thing as 'FM oscillators' there are oscillators, and FM is a thing you can do between two oscillators, both Myth and Synplant have FM, Myth doesn't use the FM in the training or resynthesis, it uses the bin array and msegs. Synplant does involve the FM in the resynthesis.
But I think it's far more complicated than this, it has a lot to do with the envelopes as well, and reverb" ~ Databroth

What are 'bin arrays' and how are MSEGS being used in the context of sounds for Myth? Are MSEGS not used (more) in sequencing contexts? In any case, in both those contexts, it would seem the sounds are being granularized. Or, perhaps the sound is being played almost per-partial, but not quite-- more like a wavetable (AKA bin array?), maybe a little like that Tomophone 'synth'?

Anyway, again, would you or anyone reading this know of where Myth's manual might be so I can RTFM?

Thanks!

Statistics: Posted by Borbolactic — Fri May 03, 2024 3:31 am

---