Without offset filter interval analysis is not truly in two voices.

[alexandermorgan]

When analyzing music with 3 or more voices and without an offset filter, if voices 1 and 2 are being analyzed, a note event in voice 3 seems to force an event in the analysis of voices 1 and 2 even at times when neither voice 1 nor voice 2 have events.

[crantila]

Could you elaborate, with an example of "expected" and "actual" results? I don't really understand what the issue is. Thanks!

[lmok]

I think we put this down to the dissonance indexer requiring forward filling. NaNs coming from instances where a voice pair don't have an event (at a point where a third voice has an event) are overwritten by a previous event. Here https://github.com/ELVIS-Project/vis/blob/diss_sigs/vis/scripts/diss_sigs.py#L75, from here https://github.com/ELVIS-Project/vis/blob/diss_sigs/vis/analyzers/indexers/dissonance.py#L41.

[alexandermorgan]

Yes, Lillio summed it up well. Sorry the original issue wasn't so clear. It seems that by not forward filling intervals and then adjusting the way the interval indexer's output dataframe is looped over in the dissonance indexer this issue should be resolved. I'm working on that now at the moment but I'm on vacation until Wednesday.

[crantila]

I can't quite remember what we decided to do about this?

[alexandermorgan]

This was sorted out by not using the 'ffill' setting when calculating dissonances. I guess I forgot to close the issue when I worked out dissonance detection. Thanks for digging this up.