On Fri, Aug 31, 2012 at 09:47:56PM +0300, Sakari Bergen wrote: > However, the biggest problem in the question is that it doesn't consider > the time-frequency uncertainty, and the fundamental nature of time limited > signals (a time limited signal can't be band limited). > > You can not measure frequencies whose period is shorter than the > measurement data. That means that you can't measure the power at 1Hz with a > resolution better than one second. This means that the "where they are" > part of the question is not well defined.
While this is 100% correct, in practice 'the soup isn't eaten as hot as it is served' (Flemish proverb, probably has equivalents in other languages). If you analyse say 200 ms of a signal then your resolution in the frequency domain is indeed limited to something like 5 Hz. So a low frequency signal such as a bass note will have its spectrum 'smeared out', but assuming there's no other signal near in frequency, the peak of the smeared out spectrum will be in the right place, and you can still discover the musical pitch. Human hearing easily violates the 'uncertainty principle', and it can do this by making assumptions about the signal (such as the one made above). If a 50 Hz bass note is a quarter tone (1.5 Hz) out of tune, we can easily hear this even if the bass plays more than 1.5 notes per second. Ciao, -- FA A world of exhaustive, reliable metadata would be an utopia. It's also a pipe-dream, founded on self-delusion, nerd hubris and hysterically inflated market opportunities. (Cory Doctorow) _______________________________________________ Linux-audio-dev mailing list [email protected] http://lists.linuxaudio.org/listinfo/linux-audio-dev
