I presume that if I'm working at 44.1kHz and I export a synth at 96kHz or 193kHz, I'm effectively upsampling the synth, right? Is it worth bouncing up to a higher sample rate like this and bringing it back into the 44.1 project? I've heard a few A/B comparisons of synths upsampled like this and in some cases I don't notice a difference, in other cases I do. 
 
Yeah I know, just "try it"....but I wondered if it's something other people here do. 

similar thread http://forum.cakewalk.com/FindPost/3046633

Sharke you need to read that thread:
 
http://forum.cakewalk.com...-bigtime-m3046633.aspx
 
In short it depends.  Craig started out by comparing a virtual instrument being rendered at 44.1K compared to upsampled to 96K, renederd up there and then down sampled to 44.1 K again.  He was expecting things to be similar but found the 96K version sounded better.
 
It depends a bit on the synth and the sound it is making.  If the sound is very warm with little high overtones and it is coming from a great analog synth emulation, then maybe not so much.  But if its 'Prism' making a very complex sound with lots of activity higher up and lots of interesting detail up high, then yes the 96 K version sounded better and I heard it as more natural  (smoother, less of it but more worthwhile)  top end that was better balanced.  ('Prism' is still producing activity as high as 48KHz!)
 
Now this is also assuming you don't put any VST's into higher precison made which some can do.  Some can internally up sample and do their thing and come out at the normal sampling rate. Put them into that mode prior to mixing and render that way for best results if they can. If the VST cannot do it then you have the option of creating a synth render session at 96K and render at that rate instead.  (You transfer the midi tracks from the 44.1K session and create a new session at 96K. Setup the same virtual instruments and play back all the parts at the new rate. Gnerate 96K renderd versions of all synth parts. Then downsample to 44.1 kHz and insert in original session.)  Converting sampling rate down to 44.1K Hz seems to retain all the smoothness and value from going up to 96K and back.
 
Meaning 44.1kHz and 16 bit is an excellent playback medium.  We just need to work at higher precisions prior to generating our final playback medium.
 

Excellent summary, Jeff. The only thing I'd add is I don't think 192kHz is worth it...then again, that's what I thought about 96kHz before I started experimenting...

Ah I forgot about that thread....I did read some of it at the time. So to be clear: I can't just export a synth track from a 44.1 session at 96kHz and have it upsampled? I would have to start a new project at 96, recreate the synth part and export it at 44.1? 
 
I do use the upsampling option of some synths and effects. Z3TA+2 has the option - sometimes I've found it sounds better, but in a couple of cases with warm 80's sounding pads it's actually sounded worse. Yeah Prism is probably a good candidate for it. 

Anderton
Excellent summary, Jeff. The only thing I'd add is I don't think 192kHz is worth it...then again, that's what I thought about 96kHz before I started experimenting...

I just bought a Babyface so for the first time the 192kHz option is on the table...doubt I'll use it though. I tend to use a lot of synths in my projects and should think things will get a little choppy fairly quickly, even though I have a decent rig. 

sharke
Ah I forgot about that thread....I did read some of it at the time. So to be clear: I can't just export a synth track from a 44.1 session at 96kHz and have it upsampled?

 
You would just be exporting a 96kHz version of a synth track that was running at 44.1kHz. You need to export the synth running at 96kHz. 
 

I do use the upsampling option of some synths and effects. Z3TA+2 has the option - sometimes I've found it sounds better, but in a couple of cases with warm 80's sounding pads it's actually sounded worse. 

Sounding better and being more accurate are not always the same thing; you might not want all those highs on warm pads. However, you'll get an even better sound if you apply a lowpass filter to the more accurate sound rather than let the computer reduce the highs for the wrong reasons, while adding in foldover distortion as a "bonus." To me, the wooliness of the foldover distortion is more problematic than the lack of highs anyway.

Anderton
 
Sounding better and being more accurate are not always the same thing; you might not want all those highs on warm pads. However, you'll get an even better sound if you apply a lowpass filter to the more accurate sound rather than let the computer reduce the highs for the wrong reasons, while adding in foldover distortion as a "bonus." To me, the wooliness of the foldover distortion is more problematic than the lack of highs anyway.

I've noticed this when using the oversampling mode of Ohmicide. I always have to use the post-effect LPF to tame the highs, and then it sounds great. 

I am a bit concerned about the process if the synth is putting out really high frequencies, without some type of filtering. Most synth documentation does not describe what the synth is doing in enough detail to say. Lets say you are using an algorithm that produces harmonics in the hypersonic range. If those frequencies were coming in via an AD converter in a sound card, they would be oversampled but primarily in order to provide a cleaner cutoff for the built in anti-aliasing filter, which will essentially eliminate those high frequencies from the signal before sampling it to the nominal sample rate. If those hypersonics are not eliminated from the signal, then the signal is not "band limited" to the audio range. One of the sometimes overlooked requirements of the Nyquist-Shannon theorem is that the signal be limited to frequencies of less than half the sampling rate in order to produce an accurate reconstruction from the samples. Preserving frequencies at higher rates before downsampling them sounds like a recipe for aliasing, which might produce a "brighter" sound to the ear without being a more accurate representation of the original signal. Of course if you are using a truly synthetic sound, the concept of fidelity may not be important, since there is no real world comparison so whatever sounds good is presumably what you want anyway. But depending on where the foldover frequencies occur, it might not only be inaccurate but unpleasant as well, and the unpleasantness might be more pronounced at certain notes than others which would produce an artifact that is not consistent across the scale. Generally a consistent sound from your synth is something that you might want. I see little to be gained by sampling your synth output at anything other than what you plan to use for the sample rate of the final recording.

One point that's being glossed over (but became apparent in the other thread) is that it is useless to upsample to 96K if the synth already has a "Hi Quality" or "2X oversampling" mode. Most do.