A point of clarification. There are two sample rates. 1) the midi sample rate by pianoteq of the digital piano. and 2) the output sample rate of the music produced by pianoteq, presumably PCM, and routed by the computer to the DAC.

Avoiding resampling is a good idea too. Windows tends to do this, I think. On linux, pipewire by default resamples everything to the default bit rate unless you configure it otherwise. I haven't yet figured out how to get pipewire on linux to output 96kHz from pianoteq. It's a bit complicated.

1. MIDI keyboard scan/transmission/buffering latency (typically 2-3ms, but can easily be more with a cheap USB interface)

2. Hidden/Unreported hardware/firmware/bus output latency of the audio interface (also typically 2-3ms and can easily be more with a cheap USB interface).

3. Speed of sound from your monitors to your ears (likely 3-4 ms unless you're wearing headphones) Of course this is no different from a real piano so can be ignored for our purposes. And if you use headphones, that can actually compensate for some of the digital latencies.

Bottom line is you can't just divide the buffer size by the sample rate and say you're getting, for example, 1.3ms of latency with a 64-sample buffer at 48kHz (or 1024 at 768kHz). In reality, the total latency from the time you press a key to the time you hear the sound is probably at least 6ms if you use headphones and 9ms if you use nearfield monitors.

If you really want to know what you're getting for latency in the digital realm, you need to do MIDI and Audio loopback tests to measure the round-trip latency of both and divide that by two (since you're only dealing with MIDI input and Audio output in the case of a soft synth).

The free CEntrance latency tester can hep with measuring ASIO round-trip latency:

https://centrance.com/driverfolder/CE_LTU_37.zip

Measuring MIDI latency requires a DAW.

mdkgr wrote:

Found a "fast" setup through ASIO4ALL also:

Try it out, but personally I've had problems with ASIO4all...

Am trying it atm. It feels the fastest (if I'm not fooling myself). No audio dropouts etc. So for now, this seems the best option before I buy a MOTU M4 or something similar. It feels kind of overkill to get a 300€ interface just to play Pianoteq but I might change my mind as the time goes on.

Found a "fast" setup through ASIO4ALL also:

Try it out, but personally I've had problems with ASIO4all...

mdkgr wrote:

dikrek wrote:

See how it feels - if exclusive mode, 48KHz and the smallest sample buffer is too slow, then probably go for a MOTU M2 (M4 if you can swing the extra cash, it’s only a tiny bit more).

If you’re very latency-sensitive then it has the best of the less expensive interfaces (and great overall performance).

Otherwise you’re going to much more expensive territory.

I'm very latency-sensitive but I can convince myself to get over it because the sound is that much better. That's what I think at the moment, at least. On long practice sessions I will have a better view...

I don't mind spending another 250. The problem is that I can't be sure what I'm getting, how different will the experience be. It's all speculative. But I feel like that I'd be satisfied.

Will try to see if I can live with what I got and then make a decision.

Try the low latency mode for windows audio. The exclusive mode for me felt higher latency. Ensure you’re at the lowest possible audio buffer size too.

So, I got three options:

The fastest is the Roland one with the exclusive audio selection. I can get used to it for sure... The low latency is the slowest amongst the three. The ASIO Chord driver is in the middle but sounds the best. I wonder how much faster would a dedicated audio interface be (like the MOTU M2).

dikrek wrote:

mdkgr wrote:

Yeah, found the option from the video that dikrek mentioned.

Currently experimenting with all the available options.

I want to find out which option offers the best experience.

But I think I can't avoid the dedicated audio interface.

Thanks again for all the inputs. You guys were SO helpful!

See how it feels - if exclusive mode, 48KHz and the smallest sample buffer is too slow, then probably go for a MOTU M2 (M4 if you can swing the extra cash, it’s only a tiny bit more).

If you’re very latency-sensitive then it has the best of the less expensive interfaces (and great overall performance).

Otherwise you’re going to much more expensive territory.

I'm very latency-sensitive but I can convince myself to get over it because the sound is that much better. That's what I think at the moment, at least. On long practice sessions I will have a better view...

I don't mind spending another 250. The problem is that I can't be sure what I'm getting, how different will the experience be. It's all speculative. But I feel like that I'd be satisfied.

Will try to see if I can live with what I got and then make a decision.

Try the low latency mode for windows audio. The exclusive mode for me felt higher latency. Ensure you’re at the lowest possible audio buffer size too.

In my experience (yet another personal opinion!), aliasing occurs more often with DA converters (poorly designed filters) than with plug-ins (unless the design is poor, here too), and, what's more, conversion from 96 to 44.1 is not free of artifacts either. Switching from 96 to 48 is of course no problem.
2 cents again :-)

Tons of plugins have horrible aliasing. Very commonly seen at 44.1. It’s just safer overall to work higher (88.2 if you want to easily convert to 44.1 even with poor converters, though any modern good converter does a perfect job converting from 96 to 44.1).

Luc Henrion wrote:

There have been many blind tests with different sampling frequencies, and almost all have come to one obvious conclusion: even "golden ears" don't pass the test. Personally, after recording in 96 KHz for a few years, I've gone back to 44.1 or 48 KHz - in 24-bit, of course, and all my customers are satisfied. This is just my opinion, but try it for yourself.

I did try it for myself and there was a difference.
Many proclaiming their ears to be golden may be mistaken and test processes can be botched.
The only test you can rely on is one that you conduct and even then, for reasons touched on by levinite, the system you test on may or may not expose differences.

There are also aliasing issues with many plugins at 40-something KHz.

Switching to 96KHz gets rid of all those problems.

So it’s not just the frequencies you can hear, it’s about artifacts audibly damaging to audio that can be avoided by going to a higher sample rate.

You can then convert the final mix from 96 to 44.1KHz and that will work great.

But the intermediate working space is 96KHz/24 bit.

There have been many blind tests with different sampling frequencies, and almost all have come to one obvious conclusion: even "golden ears" don't pass the test. Personally, after recording in 96 KHz for a few years, I've gone back to 44.1 or 48 KHz - in 24-bit, of course, and all my customers are satisfied. This is just my opinion, but try it for yourself.

I did try it for myself and there was a difference.
Many proclaiming their ears to be golden may be mistaken and test processes can be botched.
The only test you can rely on is one that you conduct and even then, for reasons touched on by levinite, the system you test on may or may not expose differences.

Without regard to latency, in theory, yes but there is a bit more than just placebo because the conversion from digital to analog determines much of the audio quality and there are a lot of reasons (filtering quality cost etc) to convert the higher sample rate signals. That conversion process involves much more than whether we hear the highest frequencies and is why we may want to spend so much on it.

Yes. When I got my first music computer, I recorded at all the different available sample rates and bit depths and while going up in sample rate produced nothing like the increase in sound quality as higher bit depth, there was a discernible difference, though nowhere near enough to justify the CPU hit and bloated file size.

mdkgr wrote:

32 or 24 bit doesn't seem to make a difference in latency but 96 or 768 kHz really does make a very noticeable difference. But as I said, I am not an expert or anything close to it.

It's just placebo effect. No human can hear above 20 kHz (that's why 44.1 kHz was chosen for CDs). In fact, most people over 40 probably will not distinguish between 48 and 32 kHz in blind tests. 44 or 48 kHz are enough fot humans.

The latencies informed by PT and other programs are not accurate, in this post there was some discussion about this:
https://forum.modartt.com/viewtopic.php?id=8170

Without regard to latency, in theory, yes but there is a bit more than just placebo because the conversion from digital to analog determines much of the audio quality and there are a lot of reasons (filtering quality cost etc) to convert the higher sample rate signals. That conversion process involves much more than whether we hear the highest frequencies and is why we may want to spend so much on it.

I don't mind spending another 250. The problem is that I can't be sure what I'm getting, how different will the experience be. It's all speculative. But I feel like that I'd be satisfied.

Will try to see if I can live with what I got and then make a decision.

If the round trip latency is as MOTU state you likely won’t feel any latency with the M2.
If it doesn’t deliver you could return it and save up some extra cash for something as fast as the Quantum.