Is it expected that Diva sounds better at higher sampling rate?

[midi_transmission]

Diva sounds great at 44.1 khz, but some sounds are 95% blindtestable better at higher sampling rates with good monitoring. Especially when having the cutoff almost open. I did not expect that clear difference actually.

With 88.1 khz it sounds more defined with clearer transients compared to 48 khz. in Devine mode the difference is smaller, but still there.

I know my basic DSP theory, it's not related to the resulting sampling rate and I verified that too. When I record a Diva part at 88.1 khz and play the wave back at 48 khz (SRC) it's still better compared to the recorded Diva part done at 48 khz. Hence, it can't be my DAC that might just work better at higher sampling rates.

Working mostly in the box, I seriously considering working at higher sampling rates with Diva now even when I don't like the consequences. It's doubling my cpi load and I will need to get a new computer sooner or later because of that.

Is this expected? And at which sampling rate do you work? Would a Devine Devine mode makes sense to overcome this with more oversampling, or is the key really to work at 88.1+ khz to have more headroom for processing in the frequency range?

[EvilDragon]

AFAIK internal DSP processing of Diva is literally clamped to 20kHz, nothing happens above that (this was done literally to reduce CPU load IIRC, but Urs will corroborate).

[midi_transmission]

I had a look at it again and there can be a measurable difference in the frequency response when you measure the spectrum. I matched with TDR Novas smart operations, which results in a about 2db boost at 18khz for the 48khz version. That could be the filter clamping at 20khz in the 48khz version?

But even after that I get 80% blind test accuracy with some patches that have a lot of high frequency and overtones, it's not only spectrum, the the sound feels a bit more defined in the high frequencies for the version created at higher sampling rate, which is not lost after downsampling the file to 48khz. Like you had a Devine^2 quality setting.

I found this very interesting, because even when the difference is pretty small, I would not hear it without a/b testing, I feel the result is even closer to good analog synth.

I ask myself if higher processing sampling rate makes sense when you want to have the last 5% of quality or if I'm chasing ghosts?

[EvilDragon]

which results in a about 2db boost at 18khz for the 48khz version. That could be the filter clamping at 20khz in the 48khz version?

Yep that's definitely the band-limiting filter doing its thing.

[zerocrossing]

I’m thinking it is time for an “Ultra Devine” mode for Diva.

[glokraw]

Maybe also a _Devina _eurorack module?

[lunardigs]

Maybe a "Divination" mode too!

[gassle]

Diva Multimode LP4 filter (modeled after Jupiter 6?) sounds clearly brighter at 88.2 khz when Cutoff is all the way up compared to 44.1. Also Osc to Osc freq modulation sounds significantly brighter or more true to my Pro one at 88.2 with Repro5. So in my experience there is definitely sound difference no matter what quality mode you're in with both plugins.

[EvilDragon]

Repro-5 is quite different from Diva, tho. Not an apples to apples comparison.

[Urs]

Quick reply:

Diva internally runs at host samplerate x fixed oversampling factor. The oversampling factor is part of the quality setting and varies between modules and models. The relation is static however, so higher host samplerates yield higher internal process rates for the same module/model. Therefore, higher samplerates get better quality and higher CPU usage.

Repro chooses its internal oversampling dynamically depending on host samplerate and quality setting. Therefore, higher sample rates yield smaller oversampling factors, and thus everything sounds pretty much the same at different samplerates, and costs same amounts of CPU, from 44.1kHz to 192kHz. Above latter, Repro behaves like Diva, but how much better sound does anyone really need?

[midi_transmission]

Thanks for chiming in and the info, Urs.

but how much better sound does anyone really need?

Disclaimer. It not about complaining or any kind of negativism! Only about reaching new heights by relatively simple tweak and thinking about whether a even higher quality mode can make sense.

I'd say for a mix with many instruments it doesn't matter.

But when you imagine hearing a song with a sparse arrangements, the synth very in focus, having a typical overtone rich analog kind of sound design, hearing it with good headphones and enjoying the sound.

I think in this situation it's worth to take advantage out of this simple tweak to process at higher sampling rate or a super high quality setting for some sounds.

As said, I can a/b it in a blind test with some sounds (not all), it's not just a theoretical thing. I feel with some sound it has a bit more definition even after matching the small frequency difference at high frequencies. At the same time, it's not that it's bad otherwise! And I could not tell the difference without an a/b comparison.

Repro chooses its internal oversampling dynamically depending on host samplerate and quality setting. Therefore, higher sample rates yield smaller oversampling factors, and thus everything sounds pretty much the same at different samplerates, and costs same amounts of CPU, from 44.1kHz to 192kHz.

Oh that's strange. I get a noticable difference in some cases between 44.1khz and 96khz.

Urs, could you check the Repro-5 preset "AZ For Sequence 1" with 44,1khz and 96khz. I think this is a rare case were the difference is easy noticeable (HQ on, MCORE on, long note f#3 at 100 vel., very noticable at 36 CUTOFF, a lot less but still with CUTOFF fully open). That said, I prefer the dirty 44.1khz sound in this case. But at 96khz it sounds smoother, clearer, maybe 'more expensive'.

I'm fully open to the idea that it's a flaw in my test when anyone has an idea.