#### Clustering around the pitch floor

Austin Zheng

I use the following steps to get a scatter plot of frequency estimates:

1. Load clips into Praat

2. Use the To Pitch (ac) program in Praat with the following parameters:
TimeStep: 60 ms;  PitchFloor: 50 Hz;  PitchCeiling: 400 Hz;  SilenceThreshold: 0;  VoicingThreshold: 0;
OctaveCost: 0;  OctaveJumpCost: 0;  VoicedUnvoicedCost: 0; Hanning window
3. Save the Pitch object in a .txt file using Praat
4. Convert the .txt file to a .csv file using Python
5. Use Python to read the .csv file, filter out all candidates with frequency above 400 Hz, and select the candidate with the highest strength for each frame. Call these the strongest candidates.
6. Create a scatter plot of the strongest candidates

When I do this, I notice that there is a cluster of frequency estimates near the pitch floor. I continue to observe this when I lower the pitch floor to 25 Hz, and then to 10 Hz. It seems to me that the clustering is a result of measurement as some function of the pitch floor, rather than the presence of noise in the clips (which should have a fixed Hz, not one that changes based on the pitch floor).

Could I ask about possible explanations for this observation?

Boersma Paul

On 7 Dec 2019, at 23:38, Austin Zheng <azaustinzheng@...> wrote:

I notice that there is a cluster of frequency estimates near the pitch floor.
The explanation is the same as the explanation of the octave cost that I answered you about before.

If you have a perfectly periodic signal with a period of 5 milliseconds, then 200 Hz is the pitch candidate that you want. However, the signal does not only repeat itself after 5 ms, but also after 10 ms, 15 ms, 20 ms, 25 ms..., so equally good pitch candidates are 100 Hz, 66.7 Hz, 50 Hz, 40 Hz...

The one that you want from all those equally good candidates, must be the highest of those frequencies; hence the octave cost parameter that I wrote about. Please look it up in the paper I wrote about this in 1993, which has all the pictures that help explain this.
_____

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/