Audio to pitch is not a solved problem. Don't expect it to work for general cases anytime soon!

From: Eirikur Hallgrimsson (eh@mad.scientist.com)
Date: Fri May 26 2000 - 00:28:31 MDT


I see a lot of people pointing optimistically to some of the roadkill
on the way to the fantasy of being able to hum and hear trumpets.
It's almost all roadkill. It'll happen eventually, but it will be
the result of using on the order of four different "pitch-guessing"
processes, and voting. And to get that to work will require
recognizing the instrument used as input. This is, after all, what a
human does. Our auditory processing is truely remarkable. The real
reason that we haven't solved this problem is that the goal has a
large fantasy element--it couldn't be made to work even using magic,
unless you somehow hacked the user's perceptions.

Assuming that you can build hardware that somehow does discriminate
pitch as fast as we do, you have the latency issue that you are now
ready to start playing the note that the human audience has already
heard from the source (whistle, guitar-string or whatever) and you
start late. You just can't make this work in real time performance
situations, even if you can hide the original control signal from the
audience. The control signal matched the timing of the other
performers, but the resulting tone is unavoidably late. Playing solo
doesn't solve it either because what you are hearing is the note that
you were humming, played as trumpets, not the one that you are trying
to hum right now. The actual note-rate (tempo) that can be
accomplished is really slow.

I'd love to see a real solution, but I don't think you can get there
even with use of EEG.

Eirikur



This archive was generated by hypermail 2b29 : Thu Jul 27 2000 - 14:11:40 MDT