On 9/17/2013 2:33 PM, Chuck Israels wrote: > Translation of audio files into notation is a more sophisticated problem, one > that requires the kind of "jumping of levels" that computers are less good > at. I can think of a number of conditions in which notation decisions reside > in areas where choices are sufficiently complex that I'd be reluctant to > trust a computer. > I quite agree with you. When you have an advanced process like this, if it is only 90% accurate, then it probably is more trouble than it is worth. This was the case of the various attempts to turn scanned sheet music into notation. Occasionally I would get something that would come through about 98% clean. That is about the break-even point I think. Anything less than that and you spend more time double-checking and fixing than you would have spent just doing it by hand in the first place.
Automatic transcribers and harmonizers might have a little more of a margin for error because I am just looking for something to get me close and I expect to do the last 5-10% myself anyway. For example, I might do drop-2 as a staring point, but then go back and move some notes around to make the voice leading smoother or make the harmonies more interesting or more characteristic for the instrument or style. It can be effective to stretch the voicing mid-phrase or to alternate between unison and ensemble, for example. If I could avoid the first 90% of the tedium, I'd be ahead of the game. That's is a weak example because I can already do some of that with the BIAB harmonization plug-in, but it really could be a whole lot smarter. Quite often I will use the BIAB harmonizer to get a first draft out for the band's first rehearsal, and then go back and do a more in-depth harmonization pass once the band has played the whole chart in context. That is the general theme of what I envision for the tools of the future: short cuts that can give us quicker results and free us to concentrate on other parts of the artistry. Having said all that, my experience with Melodyne is that it hears better than I do and is extremely accurate in its decisions. It is limited to polyphonic analysis of a single instrument track at the moment. The problem is clearly more complex in a full mix, and indeed voices that are buried have no chance of being transcribed by any program, no matter how sophisticated it gets. But fundamentally, there is no reason why a computer cannot listen to an audio file and discern the flute notes from the harp notes. If humans can hear it and easily understand the music, the computer should be able to transcribe it. This is something that would be ideal as an add-on rather an as a built-in feature because most users might not value it, but a few users would pay big bucks for it. I guess that is the point I'm getting to. There are add-on interfaces that Finale should obviously support today (e.g. Rewire, VST effects, and VST instruments). But maybe we need another industry standard for doing add-ons that are more specific to the notation process. Whereas VSTs are wave (i.e. audio file) oriented, this proposed class of add-on would be notation object-oriented -- call it VSTn or something. I don't see anybody with an appetite for this at the moment, but it wouldn't really surprise me to see that class of add-on emerge from the work that Spreadbury is doing, and I would not be surprised at all to see a very early incarnation of such an interface by 2016. _______________________________________________ Finale mailing list Finale@shsu.edu http://lists.shsu.edu/mailman/listinfo/finale