`NoveltySlice`: Unpredictable silences after parameter changes

flucoma / flucoma-core

Core algorithms and objects for the Fluid Corpus Manipulation Library

BSD 3-Clause "New" or "Revised" License

77 stars 14 forks source link

`NoveltySlice`: Unpredictable silences after parameter changes #128

Open tremblap opened 2 years ago

tremblap commented 2 years ago

if one opens the musical example tab (2) of the max helpfile (correcting the algorithm in the message to get to mfcc) one gets either nothing detected, or a continuous flow of attacks. One then needs to change the threshold to get something sensible out but it is very tempremental.

to reproduce:

open the help with DAC off, on the 2nd tab
choose nicol drum loop
play - nothing or a lot of clicks.
if nothing, press reset. usually, a lot of clicks
press the message to change preset. usually a lot of clicks. raising the threshold a little helps to tame but is strangely inconsistent.

this happened first in Pd and is tested behaviour in Max too. Beta5 and beta6. I will download beta4 and see if it is the same.

tremblap commented 2 years ago

ok I can confirm that in beta4 everything is fine and as expected. something happened between beta4 and beta5 that makes this object erratic...

weefuzzy commented 2 years ago

I'm not sure how many problems you're describing here.

Yes, I can reproduce inconsistency when there's a param change and it just shuts up
Are you also saying that those settings in the message box gave results that weren't just a click fest with nicol? Because they're pretty lively settings.

tremblap commented 2 years ago

good that you can reproduce the inconsistancy.
as for results, in beta4 it grooves, in beta5 it sounds like popcorn, for these settings. Same in SC (although I couldn't reproduce the 'silence' problem #1

weefuzzy commented 2 years ago

The first of these I've found the cause for. a4bb226e31b4d3600e348e2509ffe345bfac7e50 made something stateful but didn't reset it when needed.

The second one I'm think is a side effect of 03710d8fef6ff5dc18ceac39bacdcac789342dd0: when we fixed the internal MFCCs to run up to 20kHz on the source signal, useful thresholds would have changed. 0.018 seems ok here.

tremblap commented 2 years ago

ok good so #1 is a real bug and #2 I can fix in all help the threshold in the help