Jump to content


Please note: You can easily log in to MPN using your Facebook account!

Romplers growing up ?


Theo Verelst

Recommended Posts

Maybe some here will rather read about the latest plugin so-and-so or the latest keyboard accessory, but a lot of keyboards/workstations comparisons aren't really coming to a very definitive description of the main properties, in part because as it appears, some of the developments in the latest instruments aren't fully understood.

 

In short: lets draw line between a velocity in-sensitive instrument with a few crappy samples on one end of the line, and a well working concert grand of your favorite brand on the other side of the line, and we are interested in not just certain (obvious) quality measurements, but in the playability and sound variations.

 

One general argument could be, that it's better to have a single sample that the "keyboard player" triggers at the beginning of the song, which is actually a recording of the concert grand as it was played and recorded from the PA. Of course, that doesn't qualify as "playing".

 

The second step away from the extreme "velocity-less bad few samples" would be, obviously, to get better samples (I'm not here going into what that would have to mean exactly), add some velocity sensitivity, and make sure hitting a few notes at the same time doesn't sound all too miserable as a chord.

 

Of course most contemporary (fixed - "ROM") based sample playing machines are way beyond that, have a lot of long, decent quality samples, full normal velocity sensitivity, some effects like some resonances and stuff. Yet, it appears still hard to get a machine to do velocity cross-fades (hitting a note harder would bring in the higher velocity sample gradually, for those who would not understand the idea), and also to do perfect between-notes cross fades.

 

However, even if we have playable samples, a good amount of them, or sufficient cross fades, and there are no issues perceived with the analog output signal of such machine, for a lot of players (probably earlier for the advanced ones) it's not nice enough an instrument yet, and remains pretty far away from the other end of the line of possibilities namely the well playing concert grand.

 

Why is that? Mainly two reasons: there is a lot of sensitivity to those keys, strings and sounding board machines, and chords and repeated notes behave quite different than samples. Of course there are already some electronic/digital solutions that solve (some of) these issues. There are fully modeled digital piano imitations (though not so many new ones of other instruments in a rompler) like Pianoteq and the Roland V-Piano, there are intelligent "ROMplers" that can change the tone of notes and chords on the basis of recognizing the keys and the relative timing of the keys you play, some machines use (studio-type) of production tricks to hide velocity and chord based sounds that aren't directly possibly from normal samples, and finally there are some special machines that use advanced signal paths also for piano sounds, like the Spectral Component Modeling of the CP4.

 

As far as I can follow the latter technique, it at least changes the problem that occurs when using velocity (and possibly key-) cross fades where the cross fades interpolate the whole spectrum between the two layers, whereas it is more logical to observe the cross fade can be done per spectral (frequency) component. Also, this makes it possible to make chords sound better, without stooping down to the level where chords are mangled into some pre-determined sample soup, with no variations worthy of notion.

 

Anyhow, that's some thoughts about the subject, there are quite a few more possible.

 

T.V..

 

Link to comment
Share on other sites



  • Replies 14
  • Created
  • Last Reply

I think the Kronos was a step in the right direction, being that it does have several modeling engines on board. And it certainly has a large toolbox for designing expressive, and highly playable tones. While the PC3 series does have a solid, VA engine, and monstrous editing capabilities, it's not exactly a comprehensive workstation. Possibly the K3000, or whatever the uber-synth is called that follows the Forte, will have a working environment that is competitive with the Kronos, Fantom and Motif; we'll see.

 

Seems that the next step in workstations, if the remaining two of the big three continue in that direction, would be the integration of SCM sounds for Yamaha (and expanding that beyond pianos and ep's), and Behavior Modeling for Roland.

But considering that the Kronos was released in 2011, and nothing has hit the street since then, I'm left wondering if the age of the uber-workstation is drawing to a close.

 

(Slight diversion there; for some reason - as I was typing - I was thinking of the subject as "Workstations growing up ?". Hopefully my thoughts will still add something to the thread...)

'Someday, we'll look back on these days and laugh; likely a maniacal laugh from our padded cells, but a laugh nonetheless' - Mr. Boffo.

 

We need a barfing cat emoticon!

 

 

 

 

 

 

 

Link to comment
Share on other sites

So the best possible approach using only sampling is 128 samples per note (at every velocity) x 88 notes, all at enough length to capture it til it dies. Too big of a library for most romplers, though streaming is making such a thing more viable. But as you said, that still doesn't capture interaction between the strings/soundboard and other dynamics apart from velocity.

 

The obvious answer, which has been the answer from certain romplers, is modeling. But modeling does not dictate a HOW the modeling is implemented. I do not know the details of the Kronos engine. I know it streams a large sample library, but on top of that, you can adjust the lid position, mechanical noise, listener position, and other parameters. Some of those are likely additive, others subtractive. For instance, mechanical noise is just blending in the noise samples. Lid position would be easily implemented using a type of complex formant filter designed by comparing the frequency spectra of open vs closed positions and tailoring the response curve (filter amount) with respect to position.

 

The string interaction gets to be more complex. Theoretically, each string could have a resonance contribution from each of the up to 87 strings (well, way more than that since there are 2-3 strings per note). Theoretically, I suppose this could be mathematically modeled and harmonic content generated additively, or perhaps using FM or some other modulation. But this starts getting into some serious processing power.

 

Romplers are generally about trade-off's....using limited memory and processing power to get the most bang for the buck. Looking back over the past 30yrs, they've come a long way. How much better can they get? Time will tell, but I think as you get closer to the real thing, each increment closer will be at additional cost with diminishing returns.

Dan

 

Acoustic/Electric stringed instruments ranging from 4 to 230 strings, hammered, picked, fingered, slapped, and plucked. Analog and Digital Electronic instruments, reeds, and throat/mouth.

Link to comment
Share on other sites

One thing I've always wondered about velocity cross-fades is: don't you run the risk of phasing artifacts if two samples are playing back simultaneously?

 

Answers of 100 words or less are welcome!

 

Cheers, Mike.

Link to comment
Share on other sites

One thing I've always wondered about velocity cross-fades is: don't you run the risk of phasing artifacts if two samples are playing back simultaneously?

 

Answers of 100 words or less are welcome!

 

Cheers, Mike.

 

Of course. That is where number of samples is critical - if you only have a few, they must cross fade to be seamless. If you have many, then the change from one to the next is small, so you can do a switch rather than a cross fade. In my example, 128 samples per note would mean zero cross fading. You could do some number less than that with switching and not hear the switch points.

 

How many words was that?

Dan

 

Acoustic/Electric stringed instruments ranging from 4 to 230 strings, hammered, picked, fingered, slapped, and plucked. Analog and Digital Electronic instruments, reeds, and throat/mouth.

Link to comment
Share on other sites

Well, there's something alluring about those giga byte software samplers, if the delay is low and constant, I sure can imagine there's some fun in that.

 

But Dan is right, an intelligent workstation, even when it has been fitted with a huge Solid State Drive, will have to deal with things a bit differently, because, well the abovementioned reasons, but also, well, that's hard to say without explaining, but for chords, those samples soon become a bit useless, because as soon as waves sound through the strings and sound board, all waves will influence each other more than a bit ! I've made a thread about that a while ago, about the number of requires "chord samples" to get out of that one, and it's prohibitively high.

 

So it is interesting to search for other ways, and that has been done, I don't know exactly how in the Kronos, but from analog synthesizers to extensive products such as the Kurzweils and in terms of some resonance filtering for instance also in the Nords (not much experience here, but I've tried, listened and read).

 

The Kurzweil is I thin a special story, not just because it prepares studio stuff about working with the Equal Loudness curve, but because I presume (and to a certain extend know, see also some of my latest piano downloads for the PC3 series) there's a lot of information hidden in the samples that without processing sound a bit dated to some, which can be brought out by filtering, compression, and sub-band analysis, but that's a hard subject to summarize.

 

I've heard some of the software offerings with processing besides (big) samples, and they prove what I hold for self-evident: unless a lot of intelligent analysis went into the whole Digital Signal Processing that is in there, it's doomed to not get all to far, because of the complexity of the differential equations that govern the "real" piano's behavior !

 

T.

 

 

Link to comment
Share on other sites

I wonder if you could start with each note being a sample of that note with all other notes sounding at the same time, then apply something like a 128 band formant filter with individual band attenuations a function of what other notes are actually being played.

Dan

 

Acoustic/Electric stringed instruments ranging from 4 to 230 strings, hammered, picked, fingered, slapped, and plucked. Analog and Digital Electronic instruments, reeds, and throat/mouth.

Link to comment
Share on other sites

It would be sort of a miracle if you could extract all the correct harmonics of one piano tone and only those harmonics, and that they even have the right phase ! And, it wouldn't solve your problem of having to distinguish between single note and multiple note samples at all.

 

I have the idea it is possible to build in prepared sample correlations that will give an intelligent filter (for instance an FFT based "length" sensitive averaging filter for mid and low-mid sub=bands) information that has been craftily put in samples by using actual chords on pianos as basis, and analyzing which components are interesting to cross correlate.

 

T

 

Link to comment
Share on other sites

Well, whether additive or subtractive, the basic idea's the same....modify harmonic content based on the condition, Though I have to think there's a mathematical way to implement the interactive nature of the strings...whether it's realistic is another story.

 

So far we're talking pianos....the holy grail. So,e romplers do a fine jones modeling other instruments, particularly VA.

Dan

 

Acoustic/Electric stringed instruments ranging from 4 to 230 strings, hammered, picked, fingered, slapped, and plucked. Analog and Digital Electronic instruments, reeds, and throat/mouth.

Link to comment
Share on other sites

In Rompler-land, I suspect that a wide variety of samples/instruments are likely about as good as they're going to be: Pianos, electric pianos, bells, drums, percussion, basses, ensemble strings, ensemble woodwinds, oboes, clarinets, flutes, bassoons, pipe organs, many electric organs, and some others I am forgetting.

 

Roland and Korg have figured out that its best just to put real synths into these machines as opposed to samples of synths. Yamaha will hopefully get on board with that concept as well.

 

Opinions re: rompler Hammond emulations are usually mixed, often depending on how important those sounds are to the player. But the technology is there to produce damn realistic-sounding Hammond clones, if the manufacturer wants to choose to include things like hardware or software drawbars. I've never been much of an organist, so I am not much of a judge here.

 

Guitars remain tough, because it is so difficult to play a keyboard and have it sound like something other than a keyboardist playing guitar sounds via a keyboard. Ditto re brass sounds. But in the end, there are plenty of guitarists around, so we usually have guitarists play those sounds. And the brass patches in romplers are "good enough" for the workingman keyboardist covering brass parts off of pop records. If you need ultra-realism, you get a sophisticated sample library for your computer.

 

Roland seems to have worked some magic with a few of their latest Supernatural Acoustic sounds. But others are nothing special, and some are downright poor.

 

Bottom line - I don't foresee any earthshaking developments in Rompler-land on the horizon. But I oh, so wish that someone would surprise me.

 

Michael

Montage 8, Logic Pro X, Omnisphere, Diva, Zebra 2, etc.

http://www.youtube.com/keybdwizrd

 

Link to comment
Share on other sites

One thing I've always wondered about velocity cross-fades is: don't you run the risk of phasing artifacts if two samples are playing back simultaneously?

 

Answers of 100 words or less are welcome!

 

Cheers, Mike.

 

Yes.

 

I avoid cross-fades like the plague.

Link to comment
Share on other sites

Of course. That is where number of samples is critical - if you only have a few, they must cross fade to be seamless. If you have many, then the change from one to the next is small, so you can do a switch rather than a cross fade. In my example, 128 samples per note would mean zero cross fading. You could do some number less than that with switching and not hear the switch points.

Our survey said: 77 words

 

But narrowly beaten by:

Yes.

Thanks both, and good thread Theo.

 

Cheers, Mike.

Link to comment
Share on other sites

 

It's hard to beat a technical thread started by Theo and interesting enough to snag Dave Weiser into the conversation. :thu:

 

 

"Music expresses that which cannot be put into words and that which cannot remain silent." - Victor Hugo
Link to comment
Share on other sites

Well, whether additive or subtractive, the basic idea's the same....modify harmonic content based on the condition, ...

 

Well, you might have to be careful to stick with decent engineering logic here, Dan. It's been around since at least the 80s in wide circles to have velocity and note information directly and non-switched influencing the timbre of a sound, except with no explicit run-time chord analysis, for instance the SY99 would allow you to make very much timbre changes by velocity and key scaling all over the place (FM modulation, sample mixing, etc.), and my much used Roland Structured Adaptive Synthesis piano would also do that.

 

It's not cool to simply presume a few switching of layers complexities are good enough to hide you cannot interpolate spectrums! So without decaying into a sort of idiotic and stupid theory mastering, I maintain there's work in the interesting ew products that apparently isn't very clear yet, even to the more intelligent and (university) engineering bred crowd here, which is a shame and a big mistake, leading to all kinds of strange delusions, making it hard to even talk about "tone" and "good sound" outside of some wh*ring in the vein of "yeah great man", while in fact that is only very little bit so...

 

T.

Link to comment
Share on other sites

The way I'd attempt so solve this problem first is to have essentially a chorus effect feed into a "soundboard" resonance model (or reverb set up to mimic a soundboard). The chorus parameters would be linked to the number and values of notes being played in various ways. For instance, wet/dry would always be set so the chorus was pretty subtle, but the mix would also depend on the number of keys being held down -- more keys = more chorus.

 

The important bit would be how it's wired up; I'd do a dual-mono setup on the chorus (i.e. a separate effect for left and right) to preserve spatial relationships between the "strings" of the piano and feed that into a single "soundboard" resonator model (however fancy you want to make it).

 

Anyways, I think it's an interesting problem and the solution I just described is a relatively "cheap" approach. Another cheap trick is to track pitch by velocity within a cent or so -- repeated keystrokes on the same key then don't sound so "static".

My music http://www.cdbaby.com/Artist/Pk12

 

My Soundware (Kurzweil PC3)http://pksoundware.blogspot.com/

 

My Kurzweil PC3 Tutorials http://www.youtube.com/user/poserp.

Link to comment
Share on other sites

Archived

This topic is now archived and is closed to further replies.

×
×
  • Create New...