Alessandria: broken audio engine (96 kHz)

Problem: the polyphony randomly disappears and only a few stops and keys are working anymore. Not only during playing, also after a break when I start playing again.
Not sure whether it is related: playing fast sequences with only 1 hand drives the Polyphony and CPU indicators to their maximum in a few seconds.
Reinstalling the sample set doesn't resolve the problem.
Current solution: reset audio/MIDI

My configuration: Mac Pro Catalina, 2 x 3,46 GHz 6-Core Intel Xeon, 96 GB RAM
Hauptwerk (6.0.1) configuration:
• all stops loaded in 20-bits resolution
• sample rate 96 kHz. With 48 kHz the problem doesn't seem to occur.
• audio engine higher definition
• polyphony about 5500, when higher the audio will be broken very soon
• available memory with loaded sample set > 40 GB
• full organ disposition with coupled manuals

My other comparable sample sets (audio channels, memory usage, number of open stops) don't have this problem, even when I use the maximum polyphony level (32.768).

Questions:
• do you recognize this malbehavior specifically with the Alessandria sample set, or also with other sample sets?
• can I change something in my setup in order to avoid this behaviour?

Your help and/or suggestions will be very much appreciated!

Hi Theo - I can confirm I have noticed this behavior too. Only with Alessandria (I load in 24bit)

I am also on a Mac (mini),
3.2 GHz intel i7
Mojave OS
64 GB RAM

For me, I use 96kHz but not the pitch shifting enhancements. On my machine that gives me consistently around 4,000 polyphony.

However I have noticed that sporadically, sometimes after 30 minutes playing, and I agree - complex passage work seems to cause it to happen.

Could it be the Alessandria samples are longer (really big reverb) than many sets, and also the 6 channels per rank?

I don't have a solution - like you, reset audio/midi clears it.

Interestingly, when it happens, the CPU doesn't spike, but Hauptwerk will only play 1 or 2 notes at a time... very strange!

I will follow along this discussion, and will be happy to troubleshoot/test solutions with the group.

[mdyde]

Hello Theo/Jerry,

I don't myself have previous experience with Piotr Grabowski's Alessandria sample set ( https://piotrgrabowski.pl/alessandria/ ), although we could potentially see whether we could get a copy from Piotr if needed.

I see from his website that it's a large surround (3-perspective) organ, and it appears to be extremely wet (long reverb) from the demo video. A sample set of that size and wetness would be expected to be *extremely* CPU-intensive, especially if both the 'high-definition pitch-shifting' and 96 kHz audio engine options are turned on. I would have imagined that a 12-core Mac Pro would still be able to handle it, but I don't know for certain.

I suppose it's also conceivable that Piotr has implemented some unusual timing-sensitive key/pipe-switching logic within the sample set's organ definition that's misbehaving when the CPU is heavily loaded.

Some questions:

- You mention that the problem doesn't occur when using 48 kHz. If, as a test, on the 'Organ settings | Organ preferences | Audio engine' screen tab, you tick the 'Disable harmonic-shaping filters' option, does that prevent the problem from occurring? (A problem was found recently whereby certain deep-bass pipes could give problems with the harmonic-shaping filters at 96 kHz: viewtopic.php?f=4&t=19316&p=145127#p145095 , and I wonder whether that might be relevant.)

- Does setting your polyphony limit setting lower (e.g. to 1024, as a test), eliminate the problem?

- Are you using the Hauptwerk AU/VST Link for audio output? If so, on the 'General settings | Audio device ...) screen, if you set Hauptwerk's audio buffer size setting to exactly the same buffer size that your AU/VST host is using, and set Hauptwerk's 'number of audio buffers' setting to 1, does that eliminate the problem?

Some questions:
... test, on the 'Organ settings | Organ preferences | Audio engine' screen tab, you tick the 'Disable harmonic-shaping filters'
... setting your polyphony limit setting lower
... Are you using the Hauptwerk AU/VST Link for audio output?

Thank you so much Martin - I'll experiment this weekend and report back.

Meanwhile to your last question, this is all during realtime/live playback at the console.

I appreciate the tips!

[mdyde]

Thanks, Jerry.

Meanwhile to your last question, this is all during realtime/live playback at the console.

I assume by that you mean that you aren't using the Hauptwerk AU/VST Link. (Some people do use it for applying real-time audio effects or recording, even for live playing.)

Thanks Jerry for confirming the same problem and Martin for the great reply!

Martin, it doesn't solve the problem when I reduce the polyphony limit to 1024.
Also, I don't use the AU/VST link.

However, your suggestion to enable the checkbox 'Disable harmonic-shaping filters' seems to prevent the problem from occurring, after a couple of hours playing. CPU and polyphony indicators remain stable, independent of the complexity of the music. Said that, I have to keep the polyphony limit reduced to about 5000, increasing it activates the CPU quickly and that will mangle the output audio.

I keep testing with the checkbox activated. Also, I'm available for further testing.

[mdyde]

Thanks, Theo.

Glad that appears so far to have pinpointed it.

If it continues to work properly with the harmonic-shaping filters disabled (and with the polyphony limit set appropriately low, given that the 96 kHz and high-definition pitch-shifting options would be expected to be extremely CPU intensive), then I think it will indeed just be a symptom of the problem mentioned in that other thread ( viewtopic.php?f=4&t=19316&p=145127#p145095 ) whereby certain deep-bass pipes can cause problems with the harmonic-shaping filters. That will be addressed in the forthcoming v6.0.2, but for now you could potentially re-enable the harmonic-shaping filters and use the other work-around I mentioned there:

b) If using 96 kHz, via the voicing screen, for any affected pipes lower than about 8' C (e.g. for the bottom octave on 16' ranks, or the bottom two octaves on 32' ranks), turn:

- 'All perspectives: wind supply mod: brightness (pct)' down to zero.
- 'All perspectives: tremulant mod: brightness (pct)' down to zero.

(Or you could just stick to 48 kHz for now.)

Martin, I enabled the Harmonic-shaping filters again and used your 'work-around' for a couple of hours. During playing everything went fine. Then I left my organ for a break of about 15 minutes and when I returned the audio engine was broken again.
So, are you sure that this problem is related to the 'deep-bass' problem?

Another question:
I also have the 4-channel Martini set with the maximum polyphony limit. Although I don't exactly know the big internal differences between Alessandria and Martini, they use about the same memory. The reverb is also quite long with 6 seconds and all tremulants are sampled (which I use instead of the artificial trees).
The Martini set never shows high CPU, and even the polyphony indicator remains 0 or sometimes 1 bar.
Could that point to an internal problem of Alessandria?

Thanks again for your help!

[mdyde]

Hello Theo,

I enabled the Harmonic-shaping filters again and used your 'work-around' for a couple of hours. During playing everything went fine. Then I left my organ for a break of about 15 minutes and when I returned the audio engine was broken again.
So, are you sure that this problem is related to the 'deep-bass' problem?

Are you sure you definitely turned both of those voicing adjustments to zero for the relevant bass octaves of *all* applicable ranks? It would be very easy to miss one (especially with surround sets). Also, do likewise for all notes for *all* 'noise' ranks (including the blower, and any key-action and stop-action noises, and any other noises/effects), since they might have been implemented within the sample set as having pitches within the bass range.

If the problem definitely doesn't occur if the harmonic-shaping filters are disabled, then it will almost certainly be the problem I mentioned. (Probably those voicing adjustments hadn't been turned down to zero for some relevant virtual 'pipe', such as blower noise.)

Martin, you're right: some values weren't 0, probably due to swiping om the touchpad.
So, I checked pipe for pipe (16' lowest octave and 32' 2 lowest octaves) and added also the noises (all notes, even outside the compass). All values zero. Double checked.

Unfortunately, the audio engine broke again.

So, if I want to keep the 'high-definition pitch-shifting', I have to disable the harmonic-shaping filters. What exactly am I missing then?

Are there other settings that need to be tested, or do you need specific logs, in order to find a clue?

[mdyde]

Hello Theo,

Thanks for the update.

Are you absolutely sure that disabling the harmonic-shaping filters (via the Organ Preferences screen) does definitely stop the problem occurring (when used at 96 kHz)? (If the problem only occurs occasionally, then perhaps it was just luck that it didn't occur when the filters happened to be disabled previously.)

Also, with the filters disabled (and at 96 kHz), do you still have the very high CPU usage from the Alessandria sample set, but not (with the filters enabled) from other sample sets at 96 kHz?

If disabling the filters does fully eliminate the problem for the Alessandria at 96 kHz, then I'd be almost certain that it must be due to the low-pitch problem. I wonder whether perhaps some virtual pipes are still sounding at very low pitches (e.g. due to something unusual that Piotr might have implemented in the sample set's wind supply model), even though those pipes' default pitches (and thus note numbers on the voice screen) are higher.

[mdyde]

Hello Theo,

P.S. would you mind also using 'Help | Crate a diagnostic file' and sending the resulting file to Francois at support [at] hauptwerk.com, so that we can see whether it gives any clues?

Hi Martin,

"Are you absolutely sure that disabling the harmonic-shaping filters (...) does definitely stop the problem occurring?"
No, but I will test further with that setting checked. Also, I did reset the voicing values to their (Piotr's) defaults.

With the disabled filters it is still very easy to push the CPU ánd the polyphony to their maximum, by toggling only a few consecutive keys very quickly. With my other sample sets (96 kHz, filters enabled) this is impossible.

I sent the Diagnostic file to Francois, and I will also send a short video with the 'toggling to high CPU' exercise.

[mdyde]

Thanks, Theo.

I'll await your confirmation of whether disabling the filters eliminates the problem (even if the high CPU usage remains).

Hauptwerk doesn't do anything functionally differently at 96 kHz compared to 48 kHz (exactly the same code is executed in each case) -- the only difference is the sample rate number itself, i.e. that twice as many samples per second need to be calculated, and that pitch values are effectively correspondingly halved relative to the sample rate (so that output pitches remain correct). Aside from the filter sensitivity/stability (due to the difference in effective pitch), the only other thing relevant to the likelihood of problems occurring is performance/timing, given that 96 kHz is inherently much more CPU-intensive and timing-critical.

If the problem *does* still occur with the harmonic-shaping filters disabled, then please also try disabling the swell-box filters on the Organ Preferences screen, and see whether that eliminates the problem.

Also, are you certain that the problem doesn't occur with the sample set at 48 kHz (with all filters enabled)? Do you still get the unusually high CPU usage with it (albeit to a lesser extent, since 48 kHz is inherently half-ish as CPU-intensive as 96 kHz)?

N.B. also, if using 96 kHz you should use twice the buffer size, compared to 48 kHz, otherwise you would be making things even more timing-critical (and thus increasing the risk of performance problems). E.g. if you normally use a buffer size of 512 at 48 kHz, then you should use a buffer size of 1024 at 96 kHz, which would give the same overall latency. Also be aware that some audio drivers (e.g. MOTU's, if I recall correctly), ignore Hauptwerk's buffer size setting, so you have to change the buffer size in their control panel instead.

[mdyde]

Hello Theo/Jerry,

A quick update on this:

Theo -- thanks for the diagnostic file and video. Neither show anything that I would regard as abnormal, or give any clues to the problem you mentioned. (Your video shows the sample set using a lot of polyphony and CPU when played fast on full-organ, which both take a few seconds to die back down to zero, but that would be expected in that situation, since the sample set is extremely wet and plays 6 stereo samples simultaneously for every key press, due it being 3-perspective surround, and also layering tremulant-affected and non-tremulant-affected samples simultaneously for each pipe. 96 kHz plus high-definition pitch-shifting in that situation will of course be especially CPU-demanding.)

I've been in contact with Piotr. He said that he also sometimes experiences the problem you mentioned [whereby after playing polyphony-intensive passages, some of the pipes can no longer be played, even after leaving the organ idle for a while] in Hauptwerk v5 at 48 kHz, and also with some of his other sample sets, but more often with the Alessandria than others, given that it's more CPU-intensive.

(That being the case, the problem wouldn't be due directly to 96 kHz, or to the harmonic-shaping filters.)

He has sent us a copy of the sample set, which I'm currently downloading, to see whether I can reproduce the problem with it. Looking into that may take a few days, but I will let you know what I find.