< Back to IRCAM Forum

Antescofo Distribution 1.0-410 (210506)

Antescofo
#1

Hi,

being very conservative regarding version updates (experience taught me to be cautious) I only recently upgraded my MacBook to Catalina. At the same time I installed the recent Antescofo Max object. Mais hélas, my matured piece for flute and Antescofo-driven electronic accompaniment turned out completely broken on the first and all next tests. After sorting out quarantine issues – and improving some minor aspects regarding other IRCAM externals – it didn’t really get better, the score follower being increasingly behind the events and apparently also behind the scheduler, causing a lot of awful (at least unintended) outcome. Only after resorting to version 1.0-264 (200117) things returned to what I had composed and tried and improved.

What is the substantial difference in the recent version which can produce such macroscopic errors? Are there any directives, configuration values, practices which should be changed or added to make things work?

New Antescofo Version?
#2

Hello Kyl.
The last “official” version on the forum is v1.0-410. The previous one was v0.92. The others version are beta version and are not as stabilized as the version distributed on the Forum.

Have you the same problem with the last forum version?

The listening module is sensitive to the calibration. It is possible that the discrepancy you experienced are also due to some underlying change in your audio chains (with the OS changes, perhaps the normalization has changed also a little bit). Have you tested on the same input via some reference recording)? We are working on a better (automatic) calibration. The result will be included in the next official release.

That said, we have several reprises per year of old pieces at Ircam, and there is no problem at the exception of the adjustement that must be made to the audio input. The same remarks apply to the Metronaut application that rely on the same listening machine.

Best,
Jl.

#3

Hi JL, thank you for the considerations. To tell the truth, there are still some problems left in the run with the 1.0 beta of January 2020 under Catalina – my first examination was too superficial. Actually, at the moment I’m at a loss with it, I have to try out some things to isolate the cause or the causes. When there are findings to reason I’ll come back to the thread.

#4

Hello again, dear JL,
indeed, the remaining problems were partially caused by the signal level being to low in some sections. Above all in one which features a sequence of air pizzicati (which I had softened a bit because their level in IRCAM Solo Instruments is unrealistically loud). I could solve it by increasing the sensitivity a bit, i.e. setting gamma from the default -1.0 temporarily to -1.4. There was another problem in my use of the Max buddy object; key/value pairs were combined in a wrong order sometimes, so I learned that it is necessary to send it a “clear” message before processing a list. After updating drivers and other packages I finally obtained a decent performance of the piece, with v1.0-264 (beta) of antescofo~.mxo.

With v1.0-410, however, it still doesn’t work. Antescofo misses many notes and lags increasingly behind, so that, after a minute, the relation between events and actions is far away from that given in the score. So I still wonder what might be the difference. Could it have something to do with synchronization strategies? Though, as far as I understand it, that would not explain why so many notes are missed. Actually, I have no idea. The only difference in the set-up are the versions of the Antescofo external object.

#5

Hi Kyl,
Check that your Max/Patcher Signal Processing information also corresponds… . You might be using Antescofo (for example) with the default analysis parameters of 4096 512 but your Max might be forcing for example 1024 through its DSP Vector Size or similar.
Do you see any warnings at object instantiation in the console?
Between the versions you mention, there haven’t been any difference in the detection mecanism!!!

#6

Hi, Arshia,
yes, the Max i/o vector size is even tighter with 512 (the signal vector size is 128), because I want Max to be more reactive, while my Antescofo analysis parameters are nfft=2048, hop=256 (I also decreased nofharm from 10 to 8). In my understanding this configuration implies an overhead in filling the audio buffer and an overhead in how many chunks of samples it takes to get an fft window filled or relocated, but that should not lead to dropouts as long as the processor is fast enough (or the demand reasonable). Anyhow, with previous versions of Antescofo it worked well. Actually I had experimented a lot to find out how narrow I could go.

Well, when I increase the Max vector sizes to 2048 and 256 respectively, this should be admitted, there are less faults in the performance with v1.0-410, but only somewhat less. And in the past, as I said, it didn’t seem to be a problem that the Antescofo analysis parameters are multiples of the Max vector sizes. In the performance with v1.0-264 there are no or almost no faults using the configuration values given above.

Comparing the info output of Antescofo at different stages, I found that some values (like Gamma, Tuning etc.) now are shown with a value other than 0 (maybe because they are initialized earlier or maybe because actual values are output) – this is of course an improvement – but I didn’t find a significant difference in the actual configuration. The statistics, however, are telling, e.g. after the first 90 seconds, with v1.0-264:

Resident memory used (kB): 1013682
Virtual memory used (kB): 7458590
Total messages: 27639
Total scheduler wake-up: 11926
Total logical instants: 12090
Total inter duration less than epsilon: 25
Max length of scheduler triggered computation: 938.13 ms

And with v1.0-410:

Resident memory used (kB): 1035337
Virtual memory used (kB): 7478095
Total messages: 9988
Total scheduler wake-up: 7477
Total logical instants: 7584
Total inter duration less than epsilon: 23
Max length of scheduler triggered computation: 831.54 ms

And the corresponding history tables, with v1.0-264:

0       6436    4721    800     9       1       2       0       1       0       0        Histo Messages
1181    748     2675    5169    1837    329     10      15      65      20      31       Histo Inter logical instant (2ms)
0       0       2       34      486     494     163     791     6918    2617    421      Histo duration logical instant (0.01ms)

and with v1.0-410:

0       6405    973     125     3       0       2       0       1       0       0        Histo Messages
1570    1762    2311    943     421     505     5       8       21      8       20       Histo Inter logical instant (2ms)
1       1       8       82      372     215     889     1773    3099    762     274      Histo duration logical instant (0.01ms)

Both runs had the same configuration (the one given here at the beginning). I hope that you have an idea why this happens.
All the best, Kai Yves

#7

@giavitto, @arshiacont, any findings, insights, ideas, why the two version behave so differently on identical (pre-recorded) input? Any suggestion what I should try? I would like to adapt my score to the latest version of Antescofo but currently it seems I have to stick to the older pre-release build. All the best, Kai Yves

#8

Hello Kyl.
We can look more closely to the traces of the listenig machine. I will contact you by private email for a followup.

Thanks,
Jean-Louis.