SFZ Sample Size vis-a-vis SF2

classic Classic list List threaded Threaded
7 messages Options
Reply | Threaded
Open this post in threaded view
|

SFZ Sample Size vis-a-vis SF2

jbeach2646
My curiosity got the best of me and I found a website of a person who has developed several instruments in the SFZ format.  I download a lite version of an electric organ.  The samples folder
contains full octave samples for about 5 different stops.  I was curious and opened one of the samples in Audacity.  What I discovered was something I have known, but not used, for many
years.  The actual samples for all the instruments use only one cycle of any sound (stop voice).  You will, literally, hear nothing more than a blip when you play them.  However, instrument
(stop or voice) definition is contained in one complete cycle of a wave file.  So, they are extremely minute, literally, only .001 second and their size is in the 40-50 range.  When I first started making soundfonts in 1998, the limitation on processing speeds (75Mhz was common with 8 MB RAM), and memory led me to adopt this type of wave file editing.  My soundfonts were small
in kilobyte sizes, even though they had over 100 presets.   The beauty of SF2 is that it allows for the creation of multiple footages of instruments using one sample.   The instrument timber
is in the sample.  What I noticed with the Oboe sample of the instrument in the SFZ format which I downloaded  was that it sounded like a flute rather than an Oboe when I made an instrument of it in Polyphone.    I am not faulting the SFZ format.  However, I took an Oboe wave file and edited it to one cycle and made the exact same instrument in Polyphone.  Mine sounds like an Oboe.
 
 
One cycle samples are easy to edit and they work well in Polyphone.  What I am curious about is this.  Since the definition of a particular timber is contained in one cycle of a recording of that
instrument, why is there the perception that different frequencies or pitches of an instrument  change the  timber quality of that instrument?  This is a denial of the very concept of definition and I am trying to understand why, as this may relate to the factor of scaling in pipe ranks to, purportedly, maintain subtle aspects of timbre, there is a need to sample at different pitches when, seemingly, 16Hz and 16,000Hz, although different frequencies, may be applied to the same timber.  Its definition will be less identifiable at either 16Hz or 16,000Hz than it will be at 256Hz where its timber should be accurately identifiable.  
 
I am just a bit irate because I was criticized years ago for thinking that I could use a minute fraction of a wave file to define an instrument or organ stop.  Now, the pervasive attitude is
that SFZ is better and, yet, SFZ uses a miniscule fraction of recorded or generated sound to define a note of an instrument. 
 
John Beach

------------------------------------------------------------------------------
Check out the vibrant tech community on one of the world's most
engaging tech sites, Slashdot.org! http://sdm.link/slashdot
_______________________________________________
jOrgan-user mailing list
[hidden email]
https://lists.sourceforge.net/lists/listinfo/jorgan-user
Reply | Threaded
Open this post in threaded view
|

Re: SFZ Sample Size vis-a-vis SF2

tbeck
John, John, John.

You take one example of a bad electronic organ and assume that all sfz instruments only use tiny samples??????

Please do some more research. Or ask more questions. There is no need to be irate.

On Wed, Jun 21, 2017 at 2:08 PM, John Beach <[hidden email]> wrote:
My curiosity got the best of me and I found a website of a person who has developed several instruments in the SFZ format.  I download a lite version of an electric organ.  The samples folder
contains full octave samples for about 5 different stops.  I was curious and opened one of the samples in Audacity.  What I discovered was something I have known, but not used, for many
years.  The actual samples for all the instruments use only one cycle of any sound (stop voice).  You will, literally, hear nothing more than a blip when you play them.  However, instrument
(stop or voice) definition is contained in one complete cycle of a wave file.  So, they are extremely minute, literally, only .001 second and their size is in the 40-50 range.  When I first started making soundfonts in 1998, the limitation on processing speeds (75Mhz was common with 8 MB RAM), and memory led me to adopt this type of wave file editing.  My soundfonts were small
in kilobyte sizes, even though they had over 100 presets.   The beauty of SF2 is that it allows for the creation of multiple footages of instruments using one sample.   The instrument timber
is in the sample.  What I noticed with the Oboe sample of the instrument in the SFZ format which I downloaded  was that it sounded like a flute rather than an Oboe when I made an instrument of it in Polyphone.    I am not faulting the SFZ format.  However, I took an Oboe wave file and edited it to one cycle and made the exact same instrument in Polyphone.  Mine sounds like an Oboe.
 
 
One cycle samples are easy to edit and they work well in Polyphone.  What I am curious about is this.  Since the definition of a particular timber is contained in one cycle of a recording of that
instrument, why is there the perception that different frequencies or pitches of an instrument  change the  timber quality of that instrument?  This is a denial of the very concept of definition and I am trying to understand why, as this may relate to the factor of scaling in pipe ranks to, purportedly, maintain subtle aspects of timbre, there is a need to sample at different pitches when, seemingly, 16Hz and 16,000Hz, although different frequencies, may be applied to the same timber.  Its definition will be less identifiable at either 16Hz or 16,000Hz than it will be at 256Hz where its timber should be accurately identifiable.  
 
I am just a bit irate because I was criticized years ago for thinking that I could use a minute fraction of a wave file to define an instrument or organ stop.  Now, the pervasive attitude is
that SFZ is better and, yet, SFZ uses a miniscule fraction of recorded or generated sound to define a note of an instrument. 
 
John Beach

------------------------------------------------------------------------------
Check out the vibrant tech community on one of the world's most
engaging tech sites, Slashdot.org! http://sdm.link/slashdot
_______________________________________________
jOrgan-user mailing list
[hidden email]
https://lists.sourceforge.net/lists/listinfo/jorgan-user



------------------------------------------------------------------------------
Check out the vibrant tech community on one of the world's most
engaging tech sites, Slashdot.org! http://sdm.link/slashdot
_______________________________________________
jOrgan-user mailing list
[hidden email]
https://lists.sourceforge.net/lists/listinfo/jorgan-user
Reply | Threaded
Open this post in threaded view
|

Re: SFZ Sample Size vis-a-vis SF2

David Gritter
In reply to this post by jbeach2646

We need to be very careful distinguishing between the SFZ language, the associated sampler engines that use it, and the sample files that are used to create an instrument.   The wave files used can be as simple, or as complex and long as the instrument designer desires.  Therefore your example .wav files represent the choice made by the instrument designer, not a limitation of SFZ sampler engines, or the language.   You would probable gain greater insight into this small instrument by opening the .sfz file in a text editor to see what the designer did with his single cycle wave files.  Did he, for example, pass the oboe sample through a highly resonant filter to simulate the formant that would create a better oboe sound?  did he introduce random variations in playback of the wav?.  From my brief look at the SFZ language I have concluded the following:

SFZ will use the loop information contained in the .wav smpl chunck to create as long a loop as we might want.  The loop startpoint and endpoint can also be specified in the .sfz file.  If using the loop points in the .wav file, only the first loop will be used.
SFZ will trigger a separate release sample (needs to be an independent .wav file) when the midi note off is received.  global attack and release times will need to be properly specified in the .sfz file for this to work properly in organ simulations.

multiple releases are normally used in organ simulations to sound either the pipe turn off transient and/or the reverberation tail of the organ's environment.  thus if the release is triggered during a stacatto note when the pipe attack transient is not complete, a different release is needed.  I have not yet found how to accomplish this in SFZ commands, since the release chosen needs to be dependent on the time the main wav file has been triggered.   Alternatively GrandOrgue aFppears to offer the option of scaling the release amplitude as a function of the main file amplitude when it is stopped,  I don't see a way to do this in SFZ either.  Any help in these two areas would be appreciated.  Also, the SFZ language appears to have some sort of sequencing mechanism that will allow multiple loops to be implemented, but this also needs more study.

for a more sophisticated .sfz implementation, you might want to check out
http://www.drealm.info/sfz/DSmolken/    for a double base .sfz implementation with included .wav files

On 06/21/2017 02:08 PM, John Beach wrote:
My curiosity got the best of me and I found a website of a person who has developed several instruments in the SFZ format.  I download a lite version of an electric organ.  The samples folder
contains full octave samples for about 5 different stops.  I was curious and opened one of the samples in Audacity.  What I discovered was something I have known, but not used, for many
years.  The actual samples for all the instruments use only one cycle of any sound (stop voice).  You will, literally, hear nothing more than a blip when you play them.  However, instrument
(stop or voice) definition is contained in one complete cycle of a wave file.  So, they are extremely minute, literally, only .001 second and their size is in the 40-50 range.  When I first started making soundfonts in 1998, the limitation on processing speeds (75Mhz was common with 8 MB RAM), and memory led me to adopt this type of wave file editing.  My soundfonts were small
in kilobyte sizes, even though they had over 100 presets.   The beauty of SF2 is that it allows for the creation of multiple footages of instruments using one sample.   The instrument timber
is in the sample.  What I noticed with the Oboe sample of the instrument in the SFZ format which I downloaded  was that it sounded like a flute rather than an Oboe when I made an instrument of it in Polyphone.    I am not faulting the SFZ format.  However, I took an Oboe wave file and edited it to one cycle and made the exact same instrument in Polyphone.  Mine sounds like an Oboe.
 
 
One cycle samples are easy to edit and they work well in Polyphone.  What I am curious about is this.  Since the definition of a particular timber is contained in one cycle of a recording of that
instrument, why is there the perception that different frequencies or pitches of an instrument  change the  timber quality of that instrument?  This is a denial of the very concept of definition and I am trying to understand why, as this may relate to the factor of scaling in pipe ranks to, purportedly, maintain subtle aspects of timbre, there is a need to sample at different pitches when, seemingly, 16Hz and 16,000Hz, although different frequencies, may be applied to the same timber.  Its definition will be less identifiable at either 16Hz or 16,000Hz than it will be at 256Hz where its timber should be accurately identifiable.  
 
I am just a bit irate because I was criticized years ago for thinking that I could use a minute fraction of a wave file to define an instrument or organ stop.  Now, the pervasive attitude is
that SFZ is better and, yet, SFZ uses a miniscule fraction of recorded or generated sound to define a note of an instrument. 
 
John Beach


------------------------------------------------------------------------------
Check out the vibrant tech community on one of the world's most
engaging tech sites, Slashdot.org! http://sdm.link/slashdot


_______________________________________________
jOrgan-user mailing list
[hidden email]
https://lists.sourceforge.net/lists/listinfo/jorgan-user



------------------------------------------------------------------------------
Check out the vibrant tech community on one of the world's most
engaging tech sites, Slashdot.org! http://sdm.link/slashdot
_______________________________________________
jOrgan-user mailing list
[hidden email]
https://lists.sourceforge.net/lists/listinfo/jorgan-user
Reply | Threaded
Open this post in threaded view
|

Re: SFZ Sample Size vis-a-vis SF2

John Reimer
Administrator
David Gritter wrote
SFZ will trigger a separate release sample (needs to be an independent
.wav file) when the midi note off is received.  global attack and
release times will need to be properly specified in the .sfz file for
this to work properly in organ simulations.

multiple releases are normally used in organ simulations to sound either
the pipe turn off transient and/or the reverberation tail of the organ's
environment.  thus if the release is triggered during a stacatto note
when the pipe attack transient is not complete, a different release is
needed.  I have not yet found how to accomplish this in SFZ commands,
since the release chosen needs to be dependent on the time the main wav
file has been triggered.   Alternatively GrandOrgue aFppears to offer
the option of scaling the release amplitude as a function of the main
file amplitude when it is stopped,  I don't see a way to do this in SFZ
either.
 

David,

Thank you you this very informative post. I am very disappointed to read the final paragraph I have quoted, because it is on the issue of stacatto (or near stacatto) playing that I find playing on GO a bit problematic, if multi-releases are not being used in the particular VPO I am playing. Indications I have read on-line that GO does not handle the problem of initiating release samples particularly well, even if multiple releases are used, adds to my unease. I think that this is probably the main issue why I personally regard jOrgan as preferable to the two main alternatives. That is, I prefer to use the soundfont method of producing what I regard as satisfactory but not authentic release sounds, instead of the further complication of having to record and process multiple releases, all dependent on playing the keys for appropriately short periods of time! If the people desiring such complication spent some time trying to record pipe sounds and process them, they might soon lose their enthusiasm for it, and understand why I prefer to work with 6 samples per octave and use soundfont-generated releases!

You do not actually state that SFZ can produce release sounds in the manner we are used to with sf2, but I assume that it can, because it obviously has to be part of the sf2 specification, and I would be surprised if it is not there somewhere in the SFZ specification. As to whether SFZ does allow multiple releases to be triggered appropriately according to the length of the played note, I would certainly hope that it does, and that so far you have simply failed to see mention of it. For if it doesn't allow this, I think this debars it from the type of use we wish it to have. Surely, to adopt a positive attitude, since multiple releases have been stated to be part of its capability, we shall discover that it can, after all, allow such triggering. I want to see the use of SFZ succeed for us, because I do want jOrgan to be attractive to as many VPO users as possible, even if I personally may not feel the need for all its capabilities in any further VPO's I construct. I understand that it can play sf2 files, so that is enough for me.

On another issue, is there any information of how big a SFZ soundfont it can handle? I apologize if this information has already appeared on our Forum.

John Reimer



Reply | Threaded
Open this post in threaded view
|

Re: SFZ Sample Size vis-a-vis SF2

jbeach2646
In reply to this post by David Gritter
David G. said, “Did he, for example, pass the oboe sample through a highly resonant filter to simulate the formant that would create a better oboe sound?  did he introduce random variations in playback of the wav?. “
 
From what you say in the form of questions, I infer that the SFZ format, somehow, embeds the same type of wave/note-file parameters for articulation in playback that SF2 does.  Not having an editor to show
this in the same or similar manner to Polyphone or Vien(n)a,  leaves this to the imagination.  However, it stands to reason that sound engines must do this.  The issue seems to come down to “what can be done to something that is not of the best quality in order to improve it”  or “how good is the quality of the object (wave file) to begin with?”  We are using something, captured in time (recorded), to be played back
with time parameters, especially, if and when release samples are triggered in the sequence after a wave file’s loop for a note ceases with Note-Off. 
 
“Random variations in the playback of the wave” might be comparable to real-time, Flexible wind simulation, (vibrato) the judicious use of chorus, amounts of which can be changed or embedded to apply to the
individual wave file used for each note of a compass, or the use of MOD LFO Freq/Pitch to effectuate the subtle wavering of an Unda Maris, for example. 
 
Recordings of real pipes, on the other hand, should contain these elements from real-time recording and, since they are already part of the substance of the wave file, should, theoretically, need few or no helping parameters to define how they are to be articulated.  The difference between tremmed and untremmed samples and the two methods of stipulating the playing of each seems to be an apt analogy here.
 
Additive synthesis is starting from 0 (zero), literally, defining a timber by synthesis of harmonic partials, constituting the wave file, which is then assigned to a set of pitches/frequencies (“Play this timber (Oboe) at this Note Number (MIDI) at this Frequency/Pitch”), using stipulated parameters for articulating the sound.  These are all visible in Polyphone.
 
Given the quality of the fundamental, the wave file, it would seem that, whatever the Sound Source, Fluidsynth, Generic Sound (which, apparently, needs to be defined), or Creative Sound, which seems to be
the best source producing the best quality, audio output of SF2, the conclusion would seem to be that software synthesizers do not quite come up to the quality standard which the hardware synthesizer does.
 
By comparison and contrast, the gm.dls, used for the Microsoft General Sound Wavetable Synthesizer (Roland Sound Sample Set), while of excellent quality, does not have the dynamic quality of AWERAMGM.SF2 or many of the other General Midi SF2 files.  Whether the Direct Sound Music Producer editor (which is tedious and slow to use) could bring the gm.dls file up to equivalent quality
is a matter of speculation and would require a lot of work.  It does not allow the extent of copy-and-paste which makes instrument creation in SF2 editors a very quick procedure.
 
Forgive the digression, but it seems fitting.
 
John Beach
 
 
  

------------------------------------------------------------------------------
Check out the vibrant tech community on one of the world's most
engaging tech sites, Slashdot.org! http://sdm.link/slashdot
_______________________________________________
jOrgan-user mailing list
[hidden email]
https://lists.sourceforge.net/lists/listinfo/jorgan-user
Reply | Threaded
Open this post in threaded view
|

Re: SFZ Sample Size vis-a-vis SF2

erikds
In reply to this post by John Reimer
John Reimer wrote:
>   I have read on-line that GO does not handle the problem of
> initiating release samples particularly well, even if multiple releases are
> used,
Hello,

Would you care to tell me where you have read this information?
This is in contradiction with what i experience using e.g. the Balzan
sample set.

Thanks in advance.

Erik.

------------------------------------------------------------------------------
Check out the vibrant tech community on one of the world's most
engaging tech sites, Slashdot.org! http://sdm.link/slashdot
_______________________________________________
jOrgan-user mailing list
[hidden email]
https://lists.sourceforge.net/lists/listinfo/jorgan-user
Reply | Threaded
Open this post in threaded view
|

Re: SFZ Sample Size vis-a-vis SF2

John Reimer
Administrator
erikds wrote
Would you care to tell me where you have read this information?
This is in contradiction with what i experience using e.g. the Balzan
sample set.
Hi Erik,

I saw it in a blog of Nick Appleton, who recorded the sounds of the St Stephen's Penrith and St Augustine's Neutral Bay (?) organs, both here in the Sydney area, and if I remember correctly, the first jOrgan VPO's to use recorded samples. Paul Stratman obviously got his permission to use them in his jOrgan soundfonts.The blog is accessed via Nick's website, so it is public knowledge what his thoughts are. I think the blog may go on to describe the problem, but I don't remember for sure. I believe that Nick is on the development team of GO, but is not as involved as he was previously. I might add that his knowledge of VPO's and computer programming is probably noteworthy.

You will also find a mention of the problem in one of Colin Pykett's articles on his hugely informative website. It is in a fairly recent article where among other things he compares the use of synthesized and recorded samples.

But it has even surfaced in our own Forum, where people have commented on the strange sounds to be heard sometimes when the release sample is getting under way, especially with reed stops (I think). Put briefly, the problem is in getting the release sample to be in phase with the looped sound, during the crossfade period. There is no easy way to get this to happen, as the player may release the note at any point in the loop.

I agree with you that the problem is not always noticed, otherwise Hauptwerk and GO would both have died prematurely. I have had a quick listen to Balzan using GO, which I installed for this purpose, but have not had enough time to have a good listen. Unfortunately I am not in a position to do a good comparison, because neither Balzan nor Fontana will allow me to use the Fluidsynth backends on my system. But they do work with dsound. That won't do for the comparison, because I need to use the excellent Hibiki VST plugin to get good reverb on my system, and dsound won't allow me to do that. Those two are the only jOrgan VPO's I have tried, that exhibit that behaviour.

But to return to your observation about not hearing any problem with Balzan. I haven't heard it so far, either.

John Reimer