levinite wrote:

I did notice that Pianoteq's rendering tempo is affected by Pianoteq's midi "Playback speed" setting (slightly to the left of the midi title). Perhaps, you can use this instead of requiring the tempo modification.

That's a very interesting observation. I've never paid any attention to that setting. I just tried setting it to 120/120.096 = .9992. This caused the UI to display 1.00x intead of 1x. And the resulting render was slow/long by the expected amount.

This made me wonder if that setting might account for the issue in some cases if not this one, but it seems to default back to 1x when you restart Pianoteq.

But it could still be part of the puzzle if it's not always being properly re-initialized internally.

This is a much better fix than having to export to Sekaiju first. Thank you both!

EDIT: I can confirm that making changes to Pianoteq's "Playback speed" will change the render speed even if it does show x1.00. For my most recent test, I achieved good sync with x1.00155 for a 10-minute clip. I can also confirm that it does reset to x1 after automatically (tested by rendering without making changes after a restart), no issues there.

I'm very happy with this solution. Using the 120.096 value I got the other day should have given me 1.0008 but that was from another software on another day so maybe down to user error, or difference in hardware, etc. Anyway, I'm impressed that Pianoteq actually will reliably render the difference between 1.0016 and 1.00155. I did it twice to make sure, it's spot on.

I did notice that Pianoteq's rendering tempo is affected by Pianoteq's midi "Playback speed" setting (slightly to the left of the midi title). Perhaps, you can use this instead of requiring the tempo modification.

That's a very interesting observation. I've never paid any attention to that setting. I just tried setting it to 120/120.096 = .9992. This caused the UI to display 1.00x intead of 1x. And the resulting render was slow/long by the expected amount.

This made me wonder if that setting might account for the issue in some cases if not this one, but it seems to default back to 1x when you restart Pianoteq.

But it could still be part of the puzzle if it's not always being properly re-initialized internally.

brundlefly wrote:

Khoa wrote:

The MIDI will have a tempo of 120.00 if I record it as 120.00 in Pianoteq. Pianoteq will accurately render with whatever tempo the performance set or modified to.

So is this confirming that Pianoteq will render with correct timing if you save the performance as MIDI and re-import it vs. rendering it immediately after recording?

Unfortunately, no, simply re-importing will not fix the issue. I have to modify the tempo to 120.096 using another program (Sekaiju). But in any case, if your Roland also sends note-off and you're not seeing the issue then that theory is bust... Oh well, I'm moving on for now. My Roland is 10+ years old now so eventually it'll have to be replaced. We'll see then if the issue is still there, lol!

levinite wrote:

Hopefully, this link* can shed light on this issue.
https://majicdesigns.github.io/MD_MIDIF...iming.html

Thanks for sharing. I've considered diving down this rabbit hole to attempt to make my own simple MIDI recorder, but that will take me forever to learn haha... For now I've made my peace with it. I thought the issue was happening to everyone, then it would have been easier for Modartt to diagnose. Whereas here it's only happening to me, and maybe the original poster from a few years ago - so I don't have super high hopes. I'm still going to test the different sample rates later to see if it makes any difference.

From my reading of the older thread, the problem has never been adequately addressed. The relatively few it affects have found their own work-arounds. I wonder how many other setups would render such a long midi correctly, since we know the clock rounding is so critical. Sound cards are virtualized. As Brundlefly has suggested, it may be more an OS issue.  I did notice that Pianoteq's rendering tempo is affected by Pianoteq's midi "Playback speed" setting (slightly to the left of the midi title). Perhaps, you can use this instead of requiring the tempo modification.

Khoa wrote:

The MIDI will have a tempo of 120.00 if I record it as 120.00 in Pianoteq. Pianoteq will accurately render with whatever tempo the performance set or modified to.

So is this confirming that Pianoteq will render with correct timing if you save the performance as MIDI and re-import it vs. rendering it immediately after recording?

Unfortunately, no, simply re-importing will not fix the issue. I have to modify the tempo to 120.096 using another program (Sekaiju). But in any case, if your Roland also sends note-off and you're not seeing the issue then that theory is bust... Oh well, I'm moving on for now. My Roland is 10+ years old now so eventually it'll have to be replaced. We'll see then if the issue is still there, lol!

Thanks for sharing. I've considered diving down this rabbit hole to attempt to make my own simple MIDI recorder, but that will take me forever to learn haha... For now I've made my peace with it. I thought the issue was happening to everyone, then it would have been easier for Modartt to diagnose. Whereas here it's only happening to me, and maybe the original poster from a few years ago - so I don't have super high hopes. I'm still going to test the different sample rates later to see if it makes any difference.

So I tried another test: Using Pianoteq's virtual keyboard, I recorded a 10 minute clip with a few key presses every other minute or so. No drift! So as of now I'm pretty sure the issue is with how Pianoteq records MIDI signal from my specific digital piano (Roland HP-508). This is why no one else is able to reproduce the issue with their own hardware. And also why the issue isn't more widespread.

My current theory: The Roland HP-508 records note-off velocity, an uncommon feature according to Pianoteq (Pianoteq popped a message about it when I first calibrated my velocity curves). Could these extra events somehow be enough to add latency? I will do some more testing and see if I can disable it one way or another.

Latency will not compound over time to produce drift; If that were the case, you would just get a fixed offset of the whole performance, and generally not more than a few milliseconds. Also, Note Off Velocity is not a separate event; it's just a parameter encoded in the Note Off event that's fixed at a constant value in a keyboard that doesn't support it vs. variable from a keyboard that does. In any case, my Roland RD-700NX sends variable Note-Off Velocity, and I'm not seeing this issue. But it is interesting that you don't see it when playing fewer discrete events.

So is this confirming that Pianoteq will render with correct timing if you save the performance as MIDI and re-import it vs. rendering it immediately after recording?

(Midi Beat Time)
"Within the MIDI data stream are tempo meta-events, which contain a 24-bit value that give the number of microseconds per quarter note."

(Sequencing Time)
"The fractional number of microseconds must be properly accounted for or the MIDI playback will drift away from the correctly synchronized time."

*
https://majicdesigns.github.io/MD_MIDIF...iming.html

My current theory: The Roland HP-508 records note-off velocity, an uncommon feature according to Pianoteq (Pianoteq popped a message about it when I first calibrated my velocity curves). Could these extra events somehow be enough to add latency? I will do some more testing and see if I can disable it one way or another. 

EDIT: I'm unable to stop my digital piano from sending note-off data. Flattening the note-off curve in Pianoteq does not eliminate drift.

I suspect it's something to do with the slight difference between Mac vs Windows version of Pianoteq, just a hunch. And yes, 48,000 Hz for all.

The MIDI will have a tempo of 120.00 if I record it as 120.00 in Pianoteq. Pianoteq will accurately render with whatever tempo the performance set or modified to.

I recorded the MIDI performance simultaneously using Synthesia and Pianoteq.

If you save the Pianoteq-recorded performance to MIDI and open each of those files one a time in a DAW, does the DAW take on the same tempo in both cases?

If so, and you drop the Pianoteq-recorded MIDI file back into Pianoteq after restarting it, will it then render the proper tempo?

Khoa wrote:

Luc Henrion wrote:

As Julien wrote, Pianoteq is based on the sampling rate of the sound card used (internal or external), so in my opinion that's where you should be looking; if you could borrow an external sound card for a test, that would certainly be interesting.

Yeah, I still need to try out the different sampling rates, it's worth a shot. I'll also do some research on external sound cards. Any good ones you can recommend? Just need to be sufficient for Pianoteq. I've only used internal cards so this is new territory for me.

You have already tried 3 different laptops, i.e. 3 different soundcards. Unless you need one it might be better to wait a bit until more is known about the problem, especially since you have something of a tempo workaround.

Except that they're all internal... I suggested borrowing one for a test, not buying one; even though you can find very cheap cards that give good results - but also very expensive cards that don't!
What intrigues me is that the results can vary: too slow on the Mac, too fast on the Toshiba... very strange.
Have you checked that the Pianoteq sampling rate matches the Mac/Toshiba sampling rate? I suppose you did...

Luc Henrion wrote:

As Julien wrote, Pianoteq is based on the sampling rate of the sound card used (internal or external), so in my opinion that's where you should be looking; if you could borrow an external sound card for a test, that would certainly be interesting.

Yeah, I still need to try out the different sampling rates, it's worth a shot. I'll also do some research on external sound cards. Any good ones you can recommend? Just need to be sufficient for Pianoteq. I've only used internal cards so this is new territory for me.

You have already tried 3 different laptops, i.e. 3 different soundcards. Unless you need one it might be better to wait a bit until more is known about the problem, especially since you have something of a tempo workaround.

As Julien wrote, Pianoteq is based on the sampling rate of the sound card used (internal or external), so in my opinion that's where you should be looking; if you could borrow an external sound card for a test, that would certainly be interesting.

Yeah, I still need to try out the different sampling rates, it's worth a shot. I'll also do some research on external sound cards. Any good ones you can recommend? Just need to be sufficient for Pianoteq. I've only used internal cards so this is new territory for me.

I am still confused over your exact setup but find it curious that 23.976/24.000 is about .999 and 120.00/120.096 is approximately .9992. Is it possible the frame rate has been converted somehow?

I thought so too but I'm pretty sure it's just a coincidence. I make sure to shoot and export in 24.00 fps using DaVinci Resolve. The drift happens at the MIDI stage, though. At least that's how it is for my case.

Also, on the remote chance that it matters, what piano model, preset and rendering quaity are involved?

Some more details: Roland HP 508. Petrof Mistral Classical Recording. Rendering at 48,000 Hz, 16 bits, FLAC. High quality settings. Normalized output volume. I play in 48,000 Hz, 64 samples. 128 Polyphony. The MIDI tab is left untouched.

I'm starting to think the problem is my Roland not being fully compatible with Pianoteq somehow because I've just tried the test with 2 other laptops, one is my main M1 Macbook, the other is my old Toshiba running Windows 10. Both produced audio that don't match up. Interestingly, the Toshiba is faster, not slower.

I've imported the MIDI files into a MIDI sequencer software to see, they do drift as expected. So the cause is for sure not with "Export to audio". Maybe playing with 64 samples is too ambitious, I will try to slow it down to 128 or slower to see.