[music-dsp] Key finder

[Russell Borogove]

On Tue, Jul 27, 2004 at 07:28:03PM +0200, Didier Dambrin wrote:
> but wouldn't this result in more an 'average scale' of the score, rather
> than 'the last scale it's in'?
> 
> That is, for my case it's better to know the last scale, since it's about
> new notes that have to be added to it. If you're not resetting the scale
> detector at any time, then the scale that has the highest power might not be
> the last one, but a previous one that occured more often(?)

Yes, that's why I spoke of windowing. In my case, after adding
(all positive) weights to each scale probability, I normalized 
the scale-probabilities so the highest scored 1.0. The repeated
accumulation and normalization had the effect of decreasing the 
effective weight of older notes. It might of course just be better
to sum the weights from, say, the last 10 notes -- but if you're 
going from a busy passage to a repeated single chord (which could
belong to several different scales), you'll lose the memory of 
which scale was in effect before the repeated chord, which could 
cause the tracker to take longer to get back on target after things
got busy again. 

> And I'd probably be happy to at least know which notes would be strictly
> forbidden, that is, would sound totally dissonant. And if my brain can hear
> dissonance, there must be an algo :)

Sometimes dissonance is desirable, even in a mostly consonant 
piece.

-Borogove