rsShiftForBestMatch not working for whole waveforms

[elanhickler]

For some reason, I don't think rsShiftForBestMatch is working to align waveforms on a macro level. Instead, it seems to give a small shift value just to get the cycles in phase.

unaligned:

aligned:

comparison:

[RobinSchmidt]

in which file is this rsShiftForBestMatch function? the search function of visual studio gives me no results.

edit: ahh ok - i see - it's called rsGetShiftForBestMatch in MiscAudio.h/cpp.

edit2: haha - just reading my old comment there, what else could be done "blablabla...it's a mess..."

[RobinSchmidt]

so - what shift-value does the function actually produce in this case? can you show me a pic of that? i assume, the "aligned" and "comparison" pics are the desired results?

[RobinSchmidt]

i have an idea: apparently, the function was made for aligning some signal chunks taken somewhere from the middle of two signals...right? i don't remember - what was the original use-case? anyway - as i see it, the function rsMaxCorrelationLag that is called inside it, applies a Hamming window to both incoming signals before computing the cross-correlation. this is very likely the reason - i'd expect an effect like that from that. for chunks taken from the middle of signals, such a windowing seems a sensible thing to do (although i'm not sure, if hamming is the best choice - maybe we could experiment with other windows for this, too - edit: oh, in some other comment, i have said that i tried Hann and Blackman, too and than Hamming was indeed better - for alignment but not for cycle-marks - so more flexibility for this choice would indeed be nice).

however - for whole signals, we should perhaps use no window at all (i.e. rectangular). or, if only initial sections of samples are used, maybe a one-sided window that tapers off only to the right. hmmm...looks like we need a more flexible interface that allows client code to specify the window function to be used. ...maybe about time to consider, if a class is more appropriate than a function for this

[elanhickler]

I don't think I need this functionality after all, but I'd be willing to test if you want to work on it.

here are two files you can use to attempt to align them

two guitar samples.zip

[elanhickler]

mm actually I might still need it

[RobinSchmidt]

hmm - ok - allowing other windows is not a big change algorithm-wise, but i may need to rethink the API for that functionality...maybe a class rsCorrelator and/or rsSampleAligner or something would be convenient...