Heck yeah! Right on, that's what I'm here for.Originally Posted by Alexcremers
But we all know the second the new generation arrives, Joseph will be the first in line to diminish it.
Like you lot haven't been standing in the crowd throwing rocks before.
I believe Synful uses additive/resynthesis. That means that original sample data sets are analyzed with something like a fourier transform (wavelets, or some other form of decomposition), which extracts some partials versus time. But once these partials are obtained, it is easier, if not easy, to smoothly morph between one set to another (that is, smoothly morph between a low velocity transition to a higher velocity). Several academic methods have been worked out, and several commercial products, including Cameleon, Vertigo, Cube, Kubla? among others.
What is different about Synful is to analyze the TRANSITIONS BETWEEN TWO NOTES. And for this reason, everyone can hear something approaching realism between notes, where before there was just this "sucking sound," or at least artifacts especially created to cover up this sucking sound. However, I do wonder where Eric of Synful got the original sample sets and whether a sample developer can provide something better. I would say that the resynthesis synths named above do a poor job at acoustic simulation, even for most sustained notes. But this could simply be making CPU/analysis compromises. Certainly, there are lot more people lurking around that know a lot more about this stuff than I do.
But the bottom line is clear: Synful actually tries to analyze and model this area in time between notes. While VSL with their Legato-sample approach did something similar, VSL legato is limited to just one kind of note-note transition. There are many ways to analyze and address this note transition area. And considering how much excitement the first attempt has produced, I am hoping other developers start to address this area too. Of course, Synful, with its resynthesis morphing ability, is able to address expression changes during any time, for better vibrato control, volume swells, squeals, etc.
Very exciting times ahead, but I suspect it will still take several years.
And I agree with FredProgGH, that at the moment, Synful simply solves one problem area at the expense of others. Specific applications will want one over the other, but Synful as it exists right now, does not seem that universal of a solution.
Oh, but yes, once the engine is programmed, there is still LOTS of work involved creating the data sets, so splitting them up does make $ for time sense. Only problem is that the work has already been done for the currect crop of Synful instruments, so selling smaller collections is still discounting the work that has already been done.
Hi Guys,
The RPM Phase Database data is stored in separate .isr files per instrument.
The path is:
C:\Documents and Settings\All Users\Application Data\Synful\Synful Orchestra
These files contain the Phrase Database for each instrument as discussed on the Technology page (RPM Page) at www.synful.com.
We're not trying to be tricky by putting the data there. It's just a convenient place to put it where it can easilly be accessed by both DXi and VST plugins.
As I've mentioned the data is stored in a patented additive synthesis form. This is much more compact than PCM sampled form. That's why the files are comparatively small. However that's not the main reason why we store it that way. It is stored that way so that the phrase fragments can be easilly stretched, pitch shifted, timbre smoothed, etc.
Eric Lindemann
Thank's for the info Eric...
I have a few questions. Forgive me if they seem stupid, but I know virtually nothing about additive synthesis.
If what YBaCuO says above is correct...i.e. that there is original sample data that is analyzed in order to extract partials...then I'm guessing the final sound depends on the quality, tone and timbre of the original source from which they are extracted? Is this the case...and if it is, can the sound be "changed" either by replacing the source or alter the resolution of the extraction and/or "playback" by the Synful engine?
If partials are indeed extracted from a source, then it sounds to me that perhaps it might have some factors in common with traditional sampling (such as the quality of the source and the resolution at which it's extracted/generated).
In short...is there any room for increasing the quality of the sound by a) using a "better" source...or b) increasing extraction and/or playback resolution?
Second...I noticed that some sounds in Synful seem to be looped and that some sounds seem to change pitch over keyrange...pretty much in the same manner as a "traditional" sample would when looped and stretched.
The looping is audible in the "sustain noise" part of Synful...the loops appear short and the loop point is evident. As with the above...can these loop points be made more seamless by some sort of editing of the sustain noise?
Regarding the sustain noise and it's changing pitch over range...I might be wrong but do really the pitch of the sustain noise of real violins, violas, cellos and basses change over range?
I don't know the correct term for it (a sort of body/air resonace combined with bow friction noise I guess), but it seems to me the phenomenon of this noise changing pitch over range is usually a typical giveaway for stringfamily samplelibraries that are not recorded chromatically. If I'm right about this...can the sustain noise be changed so that it does not change pitch over range?
I can understand if you perhaps don't have the time to answer my questions...or perhaps just consider them too dumb to even bother with.
Like I said, I'm no expert (not even close)...but I find Synful very interesting and I'm curious to learn more about it and the technique on which it's based. There is no doubt in my mind that it is a revolutionary and very promising achievement...and that this is something we will see more of in the future.
Thank's
Cheers
Rodney
We still buzzing ???
So we agree, we don't like the brass ... sure.
The violin gave me a virtual Paganini complex :-)
Not sure if this has been asked yet, but is there potential to make hybrid instruments ... ones with this mystical technology combined with samples ?
This would solve the inherent problems with both camps, yes ?
Traditional samples will always serve well for percussion, do we need 'Synful' percussion, probably not.
Do we need 'Synful' brass ensembles ? I would say nope.
Do we need 'Synful' solo instruments (primarily woods and strings) ? ABSOLUTELY !!!!
So why not focus on this obvious void - not even bother with what does not need fixed.
For $495 i must have this - just for the violin. I also feel like i'm investing in the future.
Just a few thoughts.
Of couse, I would like to hear such answers to the answers above too, but we probably will not get any.
I am familiar with the Cameleon synth.
Based only on my limited understanding of Cameleon, my "answers" to the above are as follows:
Yes, there had to be real "samples" somewhere, even if these samples were created ones using various synth/mathematical methods. However, these samples are preferably completely dry. The analysis process is much more difficult when there is ambience. Cameleon was designed so that the user could import user sounds. This limitation means that the analysis part is simpler to take less time and therefore less accurate. Synful created instruments offline, so the analysis part can be more complex. However, the more complex the representation (more partials, more envelopes, etc.) the more CPU is required upon playback. Synful does seem to be very CPO friendly, and I suspect there is some good assembling coding in the critical areas, but I can only guess how "complex" the original analysis and therefore how accurate the representation is.
Cameleon was design as a synth, and really the acoustic representation is very poor. The main reason is having to have discrete frequency partials (via Fourier Transform, for example) that are separated a lot in frequency. Real acoustic instruments have complex continuous "side bands" around each frequency (well actually, the frequency spectrum is continuous), and then this has to be represented by one frequency. This causes the resynthesized sound to be very digital sounded, thin and harsh. One of the techniques to make this representation better, it is have a "noise" part, like the residue left over from the resynthesized sound. This greatly improves the representation. In Cameleon, this "noise" part then because a parameter than can be manipulated in synthesizer fashion for additional effects (usually wierd).
With this all in mind, I think that yes, quality in equals quality out. So better prepared sample sets will give better results. The real genius here; however, is choosing the right kind of transitions to create for analysis. Eric has really done a brilliant job. However, based on the results of Cameleon, and Cube, there is an inherent limitation in this kind of analysis. I am not optimistic that the sound will get that much better, except for easy instruments like the horn and flute. The violin, though, is difficult, and already Synful is getting some rave review from the violin. So maybe there is hope ... (By the way, the voice is also difficult, and there are all kinds of different analysis techniques being developed for speech synthesis that goes beyond simple Fourier Analysis. I would not be surprised if something like that is at work here - but this is all mathematical details. The general approach is the same: create the appropriate sample sets and analyze or represent them with a smaller more flexible number of parameters than are easily manipulated ...) Of course, with CPU getting 10 times more powerful soon, there can be more parameters in the representation for greater accurracy at the cost of more CPU.
Loops are easy to create in resynthesized form. Again, both the original smaple set and simple size/CPU restrictions determine the maximum loops. However, I believe that Synful loops can easily be much longer if the samples allow it. Otherwise, new samples will be needed.
As for the sustain noise changing with pitch. Again, this depends on the number of original samples used. I do not see a limitation, other than the enourmous task of sampling, editing, analyzing, etc. ...
Believe me, there are a number of lurkers who can shed much more light on this than I.
But only Eric can speak about what he plans to do with this technology in terms of marketing. He has a lot to think about: orchestral vs synthesizer; orchestral vs solo instruments, price, collaborations/selling patent licenses, etc. Here the NS community can really help. What do us consumers want?
I want more detailed, better sounding, more varieties (actual different instruments) of solo violin at a price of $150 or less.
What do other people want?
Yes the sound does depend on the original source and yes it can be replaced. Yes it can and will be possible to develop new sound libraries for the Synful Synth. The original recorded sounds are phrases from real instruments - passages from the classical literature, etc. This is explained on the technology page at www.synful.com.
eric
The first two demos were done by downloading MIDI files from the web and then tweeking the velocities, note separation/overlap, adding volume pedal and a bit of pitch bend in the Beethoven, and tempi. No mod wheel was added. Incidently by default the MOD wheel in synful is all the way on so if you never touch it, then it stays that way.
I've do alot of testing by playing live but I haven't done a full demo that way. I should do it.
Eric
Definitely. I'd like to hear a demo that represents the leading edge of what Synful can do taking advantage of all its tricks...
That's interesting background, thanks YBaCuo. So what is the difference between additive synthesis and resynthesis - are they the same thing?
Posting Permissions
Reply With Quote
Bookmarks