FAVE-extract – allowed options and parameters

  • removeStopWords:
    • True
    • False
    If True, vowels in the so-called stop words list are not measured. A basic list of stop words including prepositions and other function words (the words most likely to have reduced vowels) is included in FAVE-extract: AND, BUT, FOR, HE, HE'S, HUH, I, I'LL, I'M, IS, IT, IT'S, ITS, MY, OF, OH, SHE, SHE'S, THAT, THE, THEM, THEN, THERE, THEY, THIS, UH, UM, UP, WAS, WE, WERE, WHAT, YOU Alternatively, a list of stop words can be specified in a file uploaded by the user. The stop words file should be a simple text file, and contain exactly one stop word per line: screen shot of file of stop words
  • measureUnstressed:
    • True
    • False
    If False, vowels marked with 0 stress in the pronouncing dictionary are not measured. If True, all vowels are measured.
  • minVowelDuration:
    Any vowel with a duration shorter than this value (in seconds) will not be measured. The default setting is that vowels shorter than 50 milliseconds are not measured.
  • case:
    • upper
    • lower
    If upper, word transcriptions are output in upper case in the output file; if lower, in lower case.
  • outputFormat:
    • text
    • plotnik
    • both
    If text, the vowel formant measurements are output to a tab-delimited file. If plotnik, the output is in Plotnik format. If both, both types of files are produced as output. More information on the format of the output files can be found here.
  • outputHeader:
    • True
    • False
    If True, a header row is output as the first line of the text output file. If False, no header row is output. Only applies if the outputFormat is text.
  • formantPredictionMethod:
    • default
    • mahalanobis
    If default, the output of the default formant measurement method of the speech analysis program (Praat) is used, i.e. all vowels are measured with the same fixed set of parameters. These parameters can be adjusted using the nFormants and maxFormant options (see below, and the Praat manual for further details).
    If mahalanobis, the formant prediction algorithm from Evanini (2009) is used. This algorithm compares all poles and bandwidths returned by the LPC analysis for the vowel to a distribution of expected formant poles and bandwidths taken from the ANAE [Atlas of North American English] measurements (more details here).
  • measurementPointMethod:
    • third
    • mid
    • fourth
    • lennig
    • anae
    • faav
    This parameter determines at which point within the vowel the formant measurements are taken:
    • third measures the vowel formants at one third of the vowel's duration, for any given vowel.
    • mid measures at the vowel's midpoint, and
    • fourth at one fourth of the vowel's duration.
    • lennig uses the algorithm from Lennig (1978) to find a steady state within the vowel by minimizing the coefficient of change for F1 and F2.
    • anae uses the guidelines from the Atlas of North American English (Labov, Ash & Boberg 2006):
      • AE is measured at maximum F2
      • AO which is measured at minimum F2
      • all other vowels are measured at maximum F1
    • The default method, faav, uses the following adjustments to the third method, which were developed in the PNC project (which was originally called "FAAV"):
      • AY and EY are measured at F1 maximum (AY with an extra 25 ms padding to the left, which allows to search for the F1 maximum outside the segment boundary)
      • UW before coronal consonants (/Tuw/ in ANAE notation) is measured at the beginning of the segment
      • AW and OW are measured halfway between the beginning of the segment and maximum F1
      • all other vowels are measured one third into the vowel
  • nFormants:
    Specifies the number of formants to be returned, i.e. the order of the LPC analysis to be conducted. Only used if the formantPredictionMethod is default (see the Praat manual for further details).
  • maxFormant:
    Specifies the maximum frequency to consider for vowel formants. Only used if the formantPredictionMethod is default. Praat recommends a default value of 5000 Hz for male speakers and 5500 for females. However, adjustment may be necessary on a per-speaker basis to obtain the optimal values for this parameter, as well as for nFormants.
  • windowSize:
    In seconds, the size of the Gaussian window to be used for LPC analysis (see the Praat manual for further details).
  • preEmphasis:
    The cut-off value in Hz for the application of a 6 dB/octave low-pass filter (see the Praat manual for further details).
  • nSmoothing:
    Specifies the number of samples to be used for the smoothing of the formant tracks. The window size for the running average will be (2*nSmoothing+1). Default value is 12, which corresponds to a 25 ms window.
  • remeasurement:
    • True
    • False
    Specifies whether a second pass is performed on the data, using the speaker's own system as the base of comparison for the Mahalanobis distance. Only used if the formantPredictionMethod is mahalanobis.

Format of the options file

Instead of manually setting all of the FAVE-extract options on the web site, users can also upload their own configuration file for the FAVE-extract option. This configuration file should be a simple text file with exactly one option per line. Each line must be in the general format

option=parameter

Example:

screen shot of options file