1

I am trying to measure formant trajectories in /lV/ sequences. I started by extracting values at 11 temporal points, but I am getting a lot of incorrect values, as if Praat is struggling to recognize the first formant, which is clearly visible on the spectrogram. I am attaching a line of the obtained values, the spectrogram with the corresponding signal excerpt, and a portion of my Praat script.

        formant = noprogress To Formant (burg): 0.005, 5, 5000, 0.025, 50
        nbFormants = Get minimum number of formants
            if nbFormants >= 3
                formantTrack = noprogress Track: 3, 500, 1500, 2500, 3500, 4500, 1, 1, 1
            else
                formantTrack = formant
            endif

            duree10 = dureeCV / 10
            durationInt10 = debutsyl + (duree10 * 1)
            (...)
            durationInt90 = debutsyl + (duree10 * 9)

            select formantTrack
            f1_0 = Get value at time: 1, debutsyl, "Hertz", "Linear"
            f1_10 = Get value at time: 1, durationInt10, "Hertz", "Linear"
            (...)
            f1_90 = Get value at time: 1, durationInt90, "Hertz", "Linear"
            f1_100 = Get value at time: 1, finV, "Hertz", "Linear"

enter image description here

F1 values I get (for the whole CV interval) : 835 938 1040 1511 1605 1612 1688 1649 1720 678 1869 1484

I could remove the outliers in the next step in R, but doing so would remove a large portion of my data. Do you know any tips to obtain robust formant trajectories?

Thank you very much for any help!

1 Answer 1

0

I counted 28 F1 points for the CV interval but you report 12. So I guess you mean "the first 12 in CV". The numbers also do not track (your numbers go up, the dots go down). This suggests that your time values are wrong, that you are not getting the value at the time you think you are getting (so check that). The script-extracted formants should correspond reasonably closely to what you get with an ordinary formant listing of the formant object.

1
  • Thank you for your response, indeed, I have spotted an error in the extraction of temporal values from intervals. I also lowered the formant ceiling to 4800 Hz (I have only male adult speakers) and the values obtained seem to better match what I see on the spectrograms.
    – Jul2415
    Mar 17 at 18:18

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.