From simon@alphabeck.co.uk Tue May 18 12:03:00 2010
Subject:Re: sample over synth-Why not model and be done with it already?

To be more accurate, the New B3 (alone among its contemporaries) uses no modeling for the signal path between the digital "tonewheel" generator and the output from the keys. Each key has 10 contacts - one per drawbar and one for MIDI, so the key-click is produced in exactly the same way as in the original B3. I think the vibrato scanner may be modeled though - not sure. I have read that in principle the digital "tonewheel" generator could be simply replaced with an original mechanical one and the organ would still work correctly - it would just weigh twice as much!

Simon Beck

----- Original Message -----
From: Christian Schonberger
To: CloneWheel@yahoogroups.com
Sent: Tuesday, May 18, 2010 6:21 PM
Subject: Re: [CWSG] Re: sample over synth-Why not model and be done with it already?


Hi all,

Just my 2 cents (again) to the sample vs. synthesis question, regarding ANY instrument.

Well, I think we all can scrap the idea to sample a Hammond note by mote. It has been done, and still is in some cases - it doesn'rt work for many reasons including polyphonic behavior, phase issues, randomness of attack, release click etc. etc. etc.

So the question (regarding a Hammond) should be down to: sampling the individual tonewheels all cean, let them run freely (virtually of course) and do the entire path, drawbars, noises, key click, scanner vib, etc. and finally leslie sim through modeling. I think that's how the NI B4 (original and II), and the New B3/XK-3 series do it.

The tendency of modern multisampled instruments is the same: breaking down the samples, adding randomness and complex software architecture to make thge samples behave and controlllable to resemble the emulated instrument as closely as possible.

The problem with physical modeling is (as far as I know) that, to accurately emulate an acoustic instrument, the number of paramers and the way they interact and connect with each other is very vast. I have tried - and heard demos of - some instrument simulations which are entirely modeled (such as pianos, Hammonds, clarinet, oboe, flute, violin) and other which are made of Gigs of Samples, managed with sophisticated MIDI script or specialized engines, to closely emulate the architecture of the original instrument.

I do not count sampled phrases which are patched together. I am takling about insteruments which are playable or at least MIDI programmable to do exactly and convincingly what we want from them - and have the same expressigve range as the original.

I would say that anything using samples somehow (still) seems to have an "authentic" flavor to it, but it istn't really fully playable. Try rolls on a snare or on cymbals, try guitar strumming on a sample based guitar simulation (sounds great until you hear a real guitar with all the rich and ever-changing overtones, not to mention the X-notes, muffled, toneless strummings, and the sounds when the left hand changes position and I didn't even start with the right hnd changing position....), try coming up with a sax solo in any technique ---- you end up in trouble.
Now try anything physical modeled instrument (I guess noone still dares trying emulating a guitar or a string section this way) and the playabilityb and expresion is there. All the tiny resonances, interactions, voice stealoing, you name it - it's all there - but the overall timber still isn't right. It lacks richness and authenticity.

This in of course not so much the case with a Hammond. I don't know how the indicidual tonewheels (long loops of course) are sampled, I guess they are picked up as "early" as possible in the signal path to prevent them from being "colored" by electrical circuits. Because when you add individual tonweheels that have been sampled just using the line output signal, you have the old sampler polyphonic behavior problem. If younplay 10 Notes, you basically play 10 Hammonds, each with a different note, instead of running ten tonewheel signals through the correcty signal path with the correct tonal qualities. But if you have very pure tonewheel samples, you need to model the tonal qualities and exact routing of each tonewheel signal.

So this leads to only one question: why sample the tonewheels at all? There will be polyphonic build up of artifacts which are inaccurate. The only obstacles might be: 1) modeling the exact waveform and the exact fluctuations and 2) modeling the exact tonal qualities of the analog circuit signal path. If that can be done convincingly (I think it already IS being done convincingly in the case of Hammonds), then synthesis (= assembling from scratch, the most modern form being am exact virtual model) is the only way to go. All solutions involving samples are passing solutions to help us out until we are able to model it all.

O.K. I'm writing a book here. I apologize to everyone who fell asleep. let's have a cup of coffee, I buy ;-)

Christian


[Non-text portions of this message have been removed]



------------------------------------------------------------------------------

No virus found in this incoming message.
Checked by AVG - www.avg.com
Version: 9.0.819 / Virus Database: 271.1.1/2881 - Release Date: 05/18/10 07:26:00

[Non-text portions of this message have been removed]