Pitch floor in Praat

[zaba]

On the Praat instructional page for pitch, it says:

http://www.fon.hum.uva.nl/praat/manual/Intro_4_2__Configuring_the_pitch_contour.html

Here is why you have to supply these settings. If the pitch floor is 75 Hz, the pitch analysis method requires a 40-millisecond analysis window, i.e., in order to measure the F0 at a time of, say, 0.850 seconds, Praat needs to consider a part of the sound that runs from 0.830 to 0.870 seconds. These 40 milliseconds correspond to 3 maximum pitch periods (3/75 = 0.040). If you set the pitch floor down to 25 Hz, the analysis window will grow to 120 milliseconds (which is again 3 maximum pitch periods), i.e., all times between 0.790 and 0.910 seconds will be considered. This makes it less easy to see fast F0 changes.

So setting the floor of the pitch range is a technical requirement for the pitch analysis. If you set it too low, you will miss very fast F0 changes, and if you set it too high, you will miss very low F0 values. For children's voices you can often use 200 Hz, although 75 Hz will still give you the same time resolution as you get for the males.

But I can't make tales or heads of the first paragraph. I guess I lack a suitable technical background or else it just strikes me out of the blue.

Can someone explain what this is getting at in a way that even an undergraduate insomniac like myself could wrap his shriveled little brains around?

[MalFet]

This is related to the f0 stuff you were asking about the other day.

When Praat tries to find what it calls "Pitch", it searches an acoustic signal for repetition over a (small, in this case) period of time. Sounds with very low frequency take more time to repeat (1/frequency = period, if you remember your physics), and consequently require a larger sample window to be found.

If you use a larger sample window, you'll find the repetitions of lower frequency sounds, but you'll lose some resolution in your pitch curve.

If you use a smaller sample window, you'll have higher resolution in your pitch curve, but you'll potentially miss low frequency signals.

At least, that's always been my understanding of how Praat works. jkpate will hopefully be able to offer a more rigorous explanation.

[jkpate]

MalFet's exactly right. Lower frequencies have longer periods, so you need to wait longer to detect the repetition. To tell that a town has a farmer's market every week, you have to be there long enough to see the farmers' market repeat.

In general, when moving between time domain and frequency domain, you have a tradeoff between time and frequency resolution. If you pick a very long window for the spectrogram settings, you'll get high frequency resolution but poor time resolution.

This tradeoff is at the heart of the famous Heisenberg uncertainty principle. In quantum mechanics, particles are also waves, and a system of particles is described by a wave function. A function that is concentrated in the domain that describes things like position will be "smeared out" in its Fourier transform, which describes things like momentum.