Topics

How does pitch floor affect number of frames?

Austin Zheng
 

As far as I understand, with reference to the documentation for the ToPitch function, http://www.fon.hum.uva.nl/praat/manual/Sound__To_Pitch___.html

pitch floor will determine the length of the analysis window, and always computes 4 pitch values within one window length.

I would like to use a timestep of 0.005 and a pitch floor of 50 Hz to analyze 3 second clip. According to the documentation, each analysis window will then be 3/50 = 0.06 seconds long. Since the clip is 3 seconds long, there will be 0.06*3 = 50 analysis windows. Praat then computes 4*50 = 200 pitch values.

Yet I am also using a timestep of 0.005. According to the documentation, this should result in 3/0.005 = 600 pitch values. How do I reconcile these two results?

In addition, when I actually analyze the file in Praat, I observe that there are 589 frames. Lowering the pitch floor will result in fewer frames, and increasing the pitch floor will result in the number of frames tending towards 600. By trial and error, the relationship between pitch floor and no. of frames seems to be monotonic but nonlinear. What explains this behaviour?

Thank you.


Boersma Paul
 

On 4 Nov 2019, at 23:13, azaustinzheng@... [praat-users] <praat-users-noreply@...> wrote:

As far as I understand, with reference to the documentation for the ToPitch function, http://www.fon.hum.uva.nl/praat/manual/Sound__To_Pitch___.html
pitch floor will determine the length of the analysis window, and always computes 4 pitch values within one window length.
Only if you have an automatic window length, i.e. you supply "0" for its length.

I would like to use a timestep of 0.005 and a pitch floor of 50 Hz to analyze 3 second clip. According to the documentation, each analysis window will then be 3/50 ! = 0.06 seconds long. Since the clip is 3 seconds long, there will be 0.06*3 = 50 analysis windows. Praat then computes 4*50 = 200 pitch values.
No. If you supply a time step of 0.005 seconds explicitly, Praat will take a time step of 0.005 seconds, which will give you close to 600 frames for 3 seconds.

In addition, when I actually analyze the file in Praat, I observe that there are 589 frames. Lowering the pitch floor will result in fewer frames, and increasing th! e pitch floor will result in the number of frames tending towards 600. By trial and error, the relationship between pitch floor and no. of frames seems to be monotonic but nonlinear. What explains this behaviour?
This is one of the most asked questions. It is true of spectrograms, pitches and formants. See Intro section 3.1. The short answer is that Praat cannot know what your signal looked like before its start or after its end, so that the first pitch frame cannot lie earlier than half an analysis window after the start of the signal.
_____

Paul Boersma
Professor of Phonetic Sciences, University of Amsterdam

Visiting address: Spuistraat 134, room 632, Amsterdam
Mail: P.O. Box 1642, 1000BP Amsterdam, The Netherlands
Website: http://www.fon.hum.uva.nl/paul/