Feature Extraction

[Muhammad Talmeez]

Hi, I am working on Parkinson's disease detection. I want to extract features like

1. Noise to harmonic ratio (NHR)
2. Recurrence period density entropy (RPDE)
3. Correlation Dimension D2
4. Spread 1, Spread 2
5. and different variations of jitter and shimmer. 

 I read the documentation but can't find any help related to these. 

Can I extract these features?

[yannick...@gmail.com]

Hi!

I can help you with a few of these things, actually:
1. Praat calls HNR "Harmonicity", and Parselmouth does expose it. `Sound.to_harmonicity` should work, and give you an object you can query (`Harmonicity.get_value_at_time`).
5. Jitter and Shimmer are also available in Praat and in Parselmouth (but slightly harder to get): https://stackoverflow.com/questions/54707734/parselmouth-batch-full-voice-report. See also https://www.fon.hum.uva.nl/praat/manual/Voice_2__Jitter.html and https://www.fon.hum.uva.nl/praat/manual/Voice_3__Shimmer.html to see how Praat treats these and how you could translate them to Python.

About 2, 3, and 4, I have no idea, actually. Do you know if Praat has functionality for it, and how to access it? Parselmouth is in the first place a Python interface to Praat, so will contain all the functionality from Praat. If you know how to do it in Praat or have an existing Praat script you want to integrate into Python, I'd be happy to help you. But I don't know these measures, myself, so I would first suggest to investigate how you can get these measurements in Praat.

Kind regards

Yannick