Page 1 of 1

[Solved] [Request] Convert Recorded Samples to synth models? (if possible for the convenience of the users)

Posted: Thu Jan 11, 2018 4:19 am
by rosesbyrosex
If you have seen my previous post on this forum you may notice that i am an audio software enthusiast, im always on the latest audio plugins, trying new stuff
Recently i noticed a a growing trend of using modeled instruments that started with the launch of MODO BASS wich became a best seller in just a few days, modeled intruments are not new, some of us use partially modeled intruments (like H7S,Ample Sound Series, Some Kontakt...) and i mean partially because although the execution/playing (like strumming) of the instrument is modeled, the sound is generated via recorded samples and not models.

The most trending modeled instruments are the Physical modeled ones (i mean not just the normal modeled instrument synth, but also the physics models wich are a pain to implement [eight years took the development of the MODO BASS plugin]):
Note: there are lots of modeled instruments out there, even a modeled hurdy gurdy vst plugin, but only a few of them are Physical modeled.

Here are some advantages of Modeled Intruments over Sample Based:

SIZE ON DISK
Modeled: Small Size (5 to 10 mb average)
Sample Based: As Big the Sample Library Is The best quality (From 1 to 60 Gb average)

RAM USAGE
Modeled: Sound Models are very small programs and are being loaded on ram ( from 2 mb to 10 mb average)
Sample Based: Samples tend to be loaded on Ram so as big the sample library will also be the ram usage (from 1 to 2 gigabytes average)

DISK USAGE
Modeled: Very Small
Sample Based: Very High Specially when loading samples at startup (90% average) [REDUCES THE HARD DRIVE LIFETIME]

Variation Realism
Modeled: Automatic Model Randomization (Not repited sounds)
Sample Based: Round Robins (Wich greatly increase library size) (Samples are repited depending on the Round robin cycle length)

CPU USAGE (They are almost the same when program is loaded)
Modeled: Low Cpu just to generate audio from models.
Sample Based: some extra use of cpu usage when loading samples from disk and managing/releasing/cleaning New/Old samples from ram/system

CUSTOMIZATION
Modeled: depending on how many components the model is divided on, it can be more customized, for example pickups models, pickup selector, type of strings.. lots of possibilities models can be combined one on the top of other.
Sample Based: limited, it can be done using diferent sample banks recorded with diferent configurations like H7S Drop A feature changing the sample bank of the 7th string

ALL IN ONE INSTALLER
Modeled: Because of the small size of audio models you can include them in your main installer
Sample Based: Two individual installers, one for the sampler engine and another for the sample library

POSSIBLE SOLUTIONS?
Physical modeling isn't a easy thing (Modo Bass, Pianoteq,Swam), but models can be done in different ways, do you think that would be possible to convert the H7S sample library (9.07 gigabytes on disk) to synth models? i mean without physical modeling or individual component modeling, so it will be lightweight, and could it be done without reducing sound quality?.

Thank You Very Much

Re: Convert Recorded Samples to synth models? (if possible for the convenience of the users)

Posted: Thu Jan 11, 2018 6:47 am
by jiandingzhe
We know that the model based synthesizers are superior in several aspects: they are much smaller, and are more adjustable than sample-based instruments. However, at current we do not have enough knowledge to build a model-based virtual guitar, especially all the algorithms need to be run at real-time with very low CPU consumption.

Nevertheless, we still used some techniques to modulate the raw samples. For instance, the harmonics are dynamically tuned according to playing velocity and pitch shift. These techniques produced more RRs and pitches than the number of actually recorded ones, and reduced the number of samples to ~ 1/10.

In near future, we are going to release two more virtual guitars. They will use more advanced DSP technologies, and we are planning to further reduce the number of RRs and the density of recorded samples among pitches, in exchange of smaller sample size and more performances.

Re: Convert Recorded Samples to synth models? (if possible for the convenience of the users)

Posted: Mon Mar 19, 2018 6:35 pm
by Azundris
(removed, found my concerns addressed in another thread.)

Re: Convert Recorded Samples to synth models? (if possible for the convenience of the users)

Posted: Mon Mar 19, 2018 7:51 pm
by DarkFlameSquirrel
jiandingzhe wrote:In near future, we are going to release two more virtual guitars. They will use more advanced DSP technologies, and we are planning to further reduce the number of RRs and the density of recorded samples among pitches, in exchange of smaller sample size and more performances.


It is very exciting to hear that the already existing tech of H7S will be further expanded on, as amazing as it already is. However the RR reduction is very concerning to hear, as someone who has used many VSTi guitars for a long time.
The biggest problem that I've found in E. GTR virtual instruments is that they are almost always lacking in enough RR to sound realistic enough, mostly in a rhythm and doubletracked setting. Good examples of this are Orange Tree's E. GTR libraries. They sound terribly mechanical. Certain "tricks" used to alleviate the shortcomings of small RR don't end up working too well. I've always been a huge fan of Impact Soundworks' "Shreddage 2" line... but even those guitars that I consider some of the best in this market have some RR issues as well.
While H7S is the first E. GTR library I've seen that has broken through with some truly innovative and impressive tech, I'm still skeptical about whether or not reducing RR will really improve the sound. I have a gut feeling it can be done, but I'm not ready to believe the tech has gotten this far just yet.

Basically, I have trust issues about RR reduction due to its bad stigma and I really don't want this RR reduction plan to compromise realism. I hope for the best for these new guitars.

Re: Convert Recorded Samples to synth models? (if possible for the convenience of the users)

Posted: Thu Mar 22, 2018 6:18 am
by jiandingzhe
DarkFlameSquirrel wrote:
jiandingzhe wrote:In near future, we are going to release two more virtual guitars. They will use more advanced DSP technologies, and we are planning to further reduce the number of RRs and the density of recorded samples among pitches, in exchange of smaller sample size and more performances.


It is very exciting to hear that the already existing tech of H7S will be further expanded on, as amazing as it already is. However the RR reduction is very concerning to hear, as someone who has used many VSTi guitars for a long time.
The biggest problem that I've found in E. GTR virtual instruments is that they are almost always lacking in enough RR to sound realistic enough, mostly in a rhythm and doubletracked setting. Good examples of this are Orange Tree's E. GTR libraries. They sound terribly mechanical. Certain "tricks" used to alleviate the shortcomings of small RR don't end up working too well. I've always been a huge fan of Impact Soundworks' "Shreddage 2" line... but even those guitars that I consider some of the best in this market have some RR issues as well.
While H7S is the first E. GTR library I've seen that has broken through with some truly innovative and impressive tech, I'm still skeptical about whether or not reducing RR will really improve the sound. I have a gut feeling it can be done, but I'm not ready to believe the tech has gotten this far just yet.

Basically, I have trust issues about RR reduction due to its bad stigma and I really don't want this RR reduction plan to compromise realism. I hope for the best for these new guitars.

Your concerning is reasonable. Actually this is one of major problems we are currently concerning about, especially balancing with number of velocity/tone layers and file size.
As the variation of RR is mostly contained in its beginning part, we are trying to make a semi-synthesized instrument that latter part of sample is make by modulated sine waves. This would allow us to carry much more number of samples, as only the first ~0.5 second is needed and each sample is much smaller.

Re: Convert Recorded Samples to synth models? (if possible for the convenience of the users)

Posted: Thu Mar 22, 2018 9:13 am
by rosesbyrosex
jiandingzhe wrote:As the variation of RR is mostly contained in its beginning part, we are trying to make a semi-synthesized instrument that latter part of sample is make by modulated sine waves. This would allow us to carry much more number of samples, as only the first ~0.5 second is needed and each sample is much smaller.


that's what I mean with "synth models" [samples replaced by modulation], good to know you are already using that kind of technology, maybe in the near future you will be able to modulate the full sound of the guitar and samples will be a thing of the past.

Re: [Solved] [Request] Convert Recorded Samples to synth models? (if possible for the convenience of the users)

Posted: Wed Oct 24, 2018 2:46 am
by rosesbyrosex
Examples of products that use samples to model conversion (Sample Modeling), no physics involved.

Acoustic Samples VTines - SAMPLING MEETS MODELING https://www.acousticsamples.net/vtines

We have always had the idea that the future of virtual instruments was going to be the sound quality of sampling but with the playability of modeling. Getting both with only samples requires too many of them and only with modeling requires too much CPU.

We found a way to combine both technologies by using samples as the sound source and all the rest uses modeling techniques. This allows the V Tines to weight only 80Mb (retina graphics and presets included), to really sound like a Fender Rhodes© with all of its imperfections and to have a fully continuous response (if you're not convinced, listen to the 127 velocity example in our video) as well as a highly tweakable sound.