Loading ...
Sorry, an error occurred while loading the content.
 

Expand Messages
  • Daniel Hirst
    ... I believe the problem is not with the algorithm but with the arguments you provide it - I assume that like many users you have using the standard
    Message 1 of 1 , Dec 26, 2007
      Fernando Orphão de Carvalho said:

      > The tool for displaying glottal pulses and estimating mean pitch
      > is assigning too high pitch values (appr. 250 Hz) for segments with
      > creaky voice where, from the waveform alone, it is easy to see that
      > F0 is actually much lower.

      I believe the problem is not with the algorithm but with the
      arguments you provide it - I assume that like many users you have
      using the "standard" arguments for the pitch detection (ie 0.0 75
      600). These are hardly ever the best arguments to use.

      We have found that a good approach is to do pitch detection in two
      steps. In the first step you use standard parameters and then from
      the distribution of pitch values, you get the 1st and 3rd quartile
      which we have found are quite well correlated with the minimum and
      maximum pitch, and finally use the estimated min and max for a second
      pitch detection. This avoids a considerable number of octave errors
      which are frequently found when using the standard arguments.

      Here is a little script which you could try which incorporates this
      idea. I have no idea if this will work with non-modal speech but I
      would be very interested to hear how it works on your data.

      best wishes

      daniel


      ------------------------------------------------------------------------
      ------------------------------------------
      #praat script
      scriptName$ = "detect_pitch.praat"
      date$ = date$()
      version$ = "2007:12:26"
      #author: Daniel Hirst
      #email: daniel.hirst@...

      #purpose: detect pitch with two passes first estimating min and max
      pitch
      #requires: one Sound object (selected)


      nSounds = numberOfSelected("Sound")
      if nSounds != 1
      pause Please select one Sound
      endif
      mySound = selected()
      To Pitch... 0.01 60 750
      q25 = Get quantile... 0 0 0.25 Hertz
      q75 = Get quantile... 0 0 0.75 Hertz
      Remove
      minPitch = q25 * 0.75
      maxPitch = q75 * 1.5
      select mySound
      To Pitch... 0.01 minPitch maxPitch
      ------------------------------------------------------------------------
      ------------------------------------------


      ------------------------------------------------------------------------
      -------------------------
      Equipe Prosodie et Représentation Formelle du Langage - salle A485
      CNRS Laboratoire Parole et Langage (UMR 6057)
      Université de Provence 29 avenue Schuman
      13621 Aix-en-Provence cedex 1, France
      tel: +33-4-4295-3628 secr: +33-4-4295-3634 fax: +33-4-4295-3788
      ------------------------------------------------------------------------
      --------------------------
      Speech Prosody list http://mailup.univ-mrs.fr/wws/info/speech-
      prosody
      Speech Prosody 2008:05:6-9 Campinas, Brazil http://www.sp2008.org
      Momel-Intsint list http://tech.groups.yahoo.com/group/momel-
      intsint
      Praat-users list http://groups.yahoo.com/
      group/praat-users
      ------------------------------------------------------------------------
      --------------------------
    Your message has been successfully submitted and will be delivered to recipients shortly.