> I am a little puzzled by what you write. Guessing at what you might be > doing, I would say that your fundamental in roughly at 300Hz. Observing a > frequency in that range for 0.3 seconds gives you about 100 oscillations. > With 100 oscillations, the accuracy of the measured frequency cannot exceed > 1%, because you might have failed to account for roughly a cycle. Whatever > your do, Fourier transformations included, suffers from this fundamental > limitation, and to get better accuracy you need either more time or a higher > frequency. > > In other words, given that 1200 x log_2(101/100) = 17, a measurement of a > frequency in the 300Hz range derived from a 0.3 sec. observation cannot > produce a result with an error smaller than about 10 cents, which is bigger > than the effects you seem to talking about, but maybe I am incorrectly > interpreting your email.
Thanks for the note. I did indeed observe that the peaks in the spectrum broadened as the observation region was decreased. However, I made the assumption that the peak centroids remained unmoved despite the broadening, and thus I thought my results unaffected. It seems the validity of my assumption can be easily checked by using a tone from an electronic device and doing the Fourier integral over many different observation times, varying both in length and start time. I'll do this in the next few days. To get on or off this list see list information at http://www.cs.dartmouth.edu/~wbc/lute-admin/index.html
