Coming Soon from Camel Audio - Alchemy!

[koolkeys]

I don't think that velocity layers would be very practical considering the number of voices required to run an additive synth, but it would be cool. And I am not even close to a programmer so I wouldn't know how much actual processing it would take to do that. But I do think I read there are 100 possible "zones per source", which sounds to me like keyboard splits.

Brent

[egbert]

I seem to remember building an instrument with 2 velocity layers and several keyboard splits in Cameleon. The change in character between soft and hard hits on percussion instruments or electric bass for example are critical to capturing their character. If you can do that, you have truly resynthesised the source.

[Ben [Camel Audio]]

That last demo is brilliant !
Looking forward to this one !

Cool - thanks - I'm glad you all thought that demo was cool - I was hoping you would like it. If anyone reading this thread is wondering what makes Alchemy different, I recommend checking it out:
http://www.camelaudio.com/music/Alchemy ... Demo02.mp3

Ben, so this thing contains 600 partials, and an additional noise source with up to 250 partials? Or am I misunderstanding.

Each source has an additive engine capable of 600 partials, plus a spectral engine (tehnically, its based on a phase vocoder, but I didn't use that term because people would be bound to confuse it with the much more common use of the term vocoder). You can't really talk about a number of partials with respect to the spectral engine - thats not how it represents things. Its a cool representation that allows you to manipulate time and pitch independently with high quality. It uses between 128 and 1024 bands. Its also good for doing noise resynthesis, which is a cool effect. Each source also has a granular engine. So you have a theoretical maximum of 4 additive engines, using up to 2400 partials in total (but you'd need a very hefty machine to run that!), plus 4 spectral engines and 4 granular engines.

I can say one thing, I haven't seen a single additive synth with resynthesis that sounds as accurate as the demo you just posted. And with 600 partials, is there even another synth with that number available? I know Cube has 512, Morphine has 128, Cameleon had 64, don't remember what the DiscoDSP synth had, but it was less than Cube, IIRC.

I think the resynthesis quality definitely stands up well to the competition. With regard to number of partials - it does have more than any other additive soft synth, as far as I know. To be honest though, its not the number of partials, its what you do with them that counts - and thats where the very high quality Alchemy additive analysis engine comes in. The other thing that is important for resynthesis is using the additive and spectral engines together. You can hear some pure additive resynthesis in that demo, but you also hear resynthesis in which the spectral engine is used to resynthesise the high frequency noise components, which makes more sense. If you just use lots of additive partials and you represent the noise parts of the sound with additive partials, then if you try to pause an instant of the sound, it'll sound wrong (this isn't a restriction of Alchemy, its a general problem with the approach of using additive partials to represent noise) - the noise part won't be noisy - it'll consist of lots of strange pitche elements. If you represent the noise part with the spectral engine, it'll sound how you want it.

What are the limits on multi-sampling and resynthesis with Alchemy? Eg can you build something with multiple samples over the range and also multiple velocity layers and apply resynthesis to the whole set to generate a patch?

Each source has up to 100 zones. These zones can use which ever resynthesis method(s) you want. So if you want a multisampled guitar using additive synthesis with 100 different zones, that are spread over different velocity layers and different key ranges, you can do that. Most of the time though, just resynthesizing one sample is enough.

I don't think that velocity layers would be very practical considering the number of voices required to run an additive synth, but it would be cool. And I am not even close to a programmer so I wouldn't know how much actual processing it would take to do that. But I do think I read there are 100 possible "zones per source", which sounds to me like keyboard splits.

Each source can't crossfade or morph between different zones - it just picks the most appropriate zone when you press a key. This helps keep CPU usage down. If you want to morph between different velocity layers, you could just load one the set of multisamples for one velocity layer into source A and one set into source B and assign velocity to the morph position between them. You can even use all 4 sources like this for four way velocity morphing if you want.

Ben

[FaX]

Wow this has really got me interested.
Which is saying something.
Great demo and the resynthesis sounds superb I might add.
Very tempting indeed.

[koolkeys]

Oh, so those zones are in either dimension, from high to low frequency, OR high to low velocity? That's REALLY cool if it's true that you can distribute them as you wish.

And of course my examples on partials were only considering the number, but it looks like you've went to the extreme with the engine, so it sounds like it will be a winner in that area. Thanks for the info!

Brent

[Hunter]

Groovy demo

[krraqk]

Each source has an additive engine capable of 600 partials, plus a spectral engine (tehnically, its based on a phase vocoder, but I didn't use that term because people would be bound to confuse it with the much more common use of the term vocoder). You can't really talk about a number of partials with respect to the spectral engine - thats not how it represents things. Its a cool representation that allows you to manipulate time and pitch independently with high quality. It uses between 128 and 1024 bands. Its also good for doing noise resynthesis, which is a cool effect. Each source also has a granular engine. So you have a theoretical maximum of 4 additive engines, using up to 2400 partials in total (but you'd need a very hefty machine to run that!), plus 4 spectral engines and 4 granular engines.

Im a bit confused about that. You mean that every one of the 4 sources (A,B,C,D) can use simultaneously additive+spectral+granular engines??

Or only one of those engines can be used in the sources?

About the cpu power, how much cpu would take i. e. a sound with 4 additive sources like are standard in C5K in a normal today cpu (i. e. an CoreDuo at 3ghz)? (being 2 cores, it would be better if you can say the % is using both cores in order to make it more clear).

I think the resynthesis quality definitely stands up well to the competition. With regard to number of partials - it does have more than any other additive soft synth, as far as I know. To be honest though, its not the number of partials, its what you do with them that counts - and thats where the very high quality Alchemy additive analysis engine comes in. The other thing that is important for resynthesis is using the additive and spectral engines together. You can hear some pure additive resynthesis in that demo, but you also hear resynthesis in which the spectral engine is used to resynthesise the high frequency noise components, which makes more sense. If you just use lots of additive partials and you represent the noise parts of the sound with additive partials, then if you try to pause an instant of the sound, it'll sound wrong (this isn't a restriction of Alchemy, its a general problem with the approach of using additive partials to represent noise) - the noise part won't be noisy - it'll consist of lots of strange pitch elements. If you represent the noise part with the spectral engine, it'll sound how you want it.

The piano and flute sounds in the demo are done with both additive and spectral engines? if so, if we want to make a representation of a real sound like that flute in Alchemy, the spectral part of the sound is analysed like the additive one or it need to be user tweaked to obtain the sound desired?

Each source has up to 100 zones. These zones can use which ever resynthesis method(s) you want. So if you want a multisampled guitar using additive synthesis with 100 different zones, that are spread over different velocity layers and different key ranges, you can do that. Most of the time though, just resynthesizing one sample is enough.

Interesting the concept to represent an instrument with only one sample. Being a resynthesised sound it should be possible, but that mean that for example we can apply velocity (or any midi message) to transform the partials of the resynthesised sound to simulate the behaviour of how a real instrument mutates with different velocities? are there templates or something similar for do that?

Each source can't crossfade or morph between different zones - it just picks the most appropriate zone when you press a key. This helps keep CPU usage down. If you want to morph between different velocity layers, you could just load one the set of multisamples for one velocity layer into source A and one set into source B and assign velocity to the morph position between them. You can even use all 4 sources like this for four way velocity morphing if you want.

Ben

That compromise sounds like the new resynth algorithm of Alchemy eats a lot of cpu, (if the quality is so good its fine), maybe when we have 10 ghz octocores that morph between zones can be implemented in order to make absolutely real representations of instruments. But for now Alchemy seems very promising.

Thanks for your explanations, reading about it Alchemy is more and more interesting with the time.

[Igro]

It will be cool synth, but waveshapers in filter section is not the perfect idea. It will bring some restriction in sound design. I thought, that there will be the possibility to create serial chain(not less than three slots) of waveshapers to get cool sounds. For now, if two slots will be reserved for waveshapers, then you stay without filters. I think, that best place for waveshapers could be in LFO-Envelope section. Just my 2 cents.

Added: thanx Ben for pointing to that in other thread. Got it:)

[egbert]

I don't think that velocity layers would be very practical considering the number of voices required to run an additive synth, but it would be cool. And I am not even close to a programmer so I wouldn't know how much actual processing it would take to do that. But I do think I read there are 100 possible "zones per source", which sounds to me like keyboard splits.

The number of voices (oscillators is what I think you meant to say) and the number of velocity layers are really quite unrelated as I understand it. The number of oscillators will be the same whether or not you are modeling the response to velocity.

For each sampled note from the source instrument included, the analysis will detect a particular a set of partials, each with its own envelope. If you admit velocity switching, then that just involves a different set of coefficients for the partials and their related envelopes for each velocity level analysed.

A resynthesis model could thus analyse a stack of velocity mapped samples and determine the partials present and their envelopes and develop a model for how velocity effects the levels and envelopes of each of the partials. Using interpolation, the resynthesis engine could use this model to mimic the tonal variation with velocity for that tone. If you also want the instrument model to also change in timbre over the pitch range of source instrument you also need samples covering that range and then your resynthesis model can mimic both change with note pitch and change with velocity.

[kritikon]

Despite all the nitpickings about what it might/might not include (rather pointless speculation IMO...), I listened to that vocal demo from Ben

http://www.camelaudio.com/music/AlchemyAnalysisDem o02.mp3

...and was rather stunned by it. I don't often get excited by synths, but this seems to be the ultimate in sampling to date. To do that type of thing with most other samplers would be either:
A) impossible
or
B) would take forever snipping the whole thing up and applying god knows how many different effects and processes to individual clips.


Was that demo all done on the fly? If so...I'll buy the thing undemoed - I'm sure it'll take me all year to figure out how it works, but that thing sounds like it does what I always hoped for samplers to do, but so far haven't done.
200 Euro...so that'd be NZ$400 - a ****ing bargain IMO. I'm sold.


Good work Camel (is your surname Toe by any chance? )

[standalone]

I'm sure it'll take me all year to figure out how it works

I guess you don't know Chameleon already , you shouldn't be worried about that ! .

[pdxindy]

That is the best sounding resynthesis I have heard to date! Very accurate! It is the first time I have heard it just sound like what it is resynthesizing, rather than that weird metallic twang it always gets. The noise in the flute was spot on!

I look forward to it being available!

[griffin]

I have just noticed the patch name on the screen shot on the alchemy page, that gave me a little giggle. This synth has me interested for sure.

[biomechanoid]

hi krraqk,

sorry posted my reply on the other thread by mistake:

krraqk wrote:

Im a bit confused about that. You mean that every one of the 4 sources (A,B,C,D) can use simultaneously additive+spectral+granular engines??

Or only one of those engines can be used in the sources?

Each source can be either granular, additive, spectral or additive+spectral. VA is a different implementation of the additive mode and can be used alone or layered with any other non-additive mode within a single source.

About the cpu power, how much cpu would take i. e. a sound with 4 additive sources like are standard in C5K in a normal today cpu (i. e. an CoreDuo at 3ghz)? (being 2 cores, it would be better if you can say the % is using both cores in order to make it more clear).

Loading the same patch from Cameleon into Alchemy should use significantly less CPU due to optimized SSE code.

Understandably, there have been several questions re CPU usage. I'll try to address them all here.

The factory sounds have all been programmed not to exceed 50% CPU of a single core 2.0ghz playing 6 notes. Many sounds will use considerably less.

Alchemy offers a lot of flexiblity, and its certainly possible to bring any system to its knees - (ref. beej's 1200 oscillator VA patch ) However all our tests to date have shown its CPU use compares favourably with comparable synths. eg - in granular mode with the same sample and settings, Alchemy is slightly more efficient than Absynth. In Additive mode Alchemy with the same sample and settings, Alchemy is slightly more efficient than Cube.

The piano and flute sounds in the demo are done with both additive and spectral engines? if so, if we want to make a representation of a real sound like that flute in Alchemy, the spectral part of the sound is analysed like the additive one or it need to be user tweaked to obtain the sound desired?

The flute was analysed with add+spect (at default settings) into a single source. If i recall correctly, the number of additive partials was reduced slightly from the default of 130, and that's all the editing involved. It's possible to edit in minute detail with the additive and spectral editors, but we think its important to show Alchemy's analysis without any extensive or specialized editing. All the sounds in the demo only had very minor tweaks made to them at Source level.

Interesting the concept to represent an instrument with only one sample. Being a resynthesised sound it should be possible, but that mean that for example we can apply velocity (or any midi message) to transform the partials of the resynthesised sound to simulate the behaviour of how a real instrument mutates with different velocities? are there templates or something similar for do that?

You can assign any modulator to control additive pitch, pan and amp... amp currently has options for brightness, fifths, multiples of 5ths, fundamental, octave and odd/even. Users or preset designers can add their own templates to these by creating a spreadsheet or text file.

Thanks

Col

[biomechanoid]

Was that demo all done on the fly?

No, to save time and make it easy we cut it into sections and programmed separate patches. With a bit of time and effort, it would be entirely possible to do most of the speech from a single, heavily automated Alchemy patch.