Problem: I get different formant values if I choose to analyse 3 formants than if I choose to analyse 4 formants.
Solution: the "number of formants" in formant analysis determines the number of peaks with which the entire spectrum is modelled. For an average female voice, you should choose to analyse 5 formants in the region up to 5500 Hz, even if you are interested only in the first three formants.
Problem: I often get only 1 formant in a region where I see clearly 2 formants in the spectrogram.
This occurs mainly in back vowels (F1 and F2 close together) for male voices, if the "maximum formant" is set to the standard of 5500 Hz, which is appropriate for female voices. Set the "maximum formant" down to 5000 Hz. No, Praat comes without a guarantee: the formant analysis is based on LPC, and this comes with several assumptions as to what a speech spectrum is like.
Question: what algorithm is used for formant analysis?
Answer: see Sound: To Formant (burg)....
© ppgb, September 16, 2003