What is FAVE-extract?

FAVE-extract is part of the FAVE (Forced Alignment & Vowel Extraction) program suite, which comprises two programs: FAVE-align and FAVE-extract.

FAVE-extract allows you to automatically extract formant measurements for a given speaker in an aligned sound file. It accepts as input a sound file with its corresponding aligned Praat TextGrid, and returns a file with the vowel measurements to the user via email.

(As with FAVE-align, FAVE-extract currently works with English-language data only.)

How to use FAVE-extract

screen shot of how to upload the FAVE-extract input files

Input files.

Upload the files to be analyzed:

  • a sound file
  • a Praat TextGrid file

Note: The uploaded Praat TextGrid must contain exactly two tiers per speaker: one phone tier, and one word tier (see image "TextGrid with multiple speakers" below). Typically, this will be the output of FAVE-align, but you can also use any another aligner using the ARPAbet phone set.

screen shot of a sample TextGrid that includes a style tier

Format of the input TextGrid and style tier. Click on the image for a larger version.

Optionally, the TextGrid may also include a style tier coding different speaking styles. This tier, if present, must be labelled "style" and be the last tier of the TextGrid. If the style codes outlined in section 3.4 of the PNC transcription guidelines are used, they will be automatically converted to the corresponding Plotnik style codes by FAVE-extract.

Supply the necessary information:

screen shot of the speaker background information fields

Speaker background information.

FAVE-extract allows you to specify all kinds of background information about the speaker to be analyzed. Most of this information is not necessary for the correct funtioning of the program and will only be included in the header of the output formant file (if your chosen options include the presence of a header in the output file).

There are, however, two fields in the FAVE-extract interface that are mandatory:

  • which speaker you want to analyse in the TextGrid (the speaker number)
  • if you are using the Mahalanobis formant prediction method: whether your speaker is male or female

What is the speaker number ("Number in TextGrid")?

screen shot of the field in which the speaker number should be entered

Speaker number input field ("Number in TextGrid"). Click on the image for a larger version.

FAVE-extract needs to know which speaker you want to analyze in a TextGrid with multiple speakers. This information is encoded in the speaker number, which corresponds to the order of the speakers from the top to the bottom of the TextGrid.

screen shot of an example TextGrid with multiple speakers

TextGrid with multiple speakers. Click on the image for a larger version.

In a TextGrid with n different speakers, there will be 2*n tiers in total: Each speaker is represented by two tiers; one phone tier and one word tier. (If the TextGrid includes a style tier, this will raise the total number of tiers to 2*n+1.)

For example, the TextGrid on the right (click on the thumbnail to see the full version of the image) contains ten tiers for four different speakers (including two tiers for the background noises): "Mary Costa", "Mr. Costa", "Interviewer", and "Interviewer 2". The speaker numbers for each of these speakers would be as follows:

Speaker name Speaker number Tier numbers
Mary Costa 1 1, 2
Mr. Costa 2 3, 4
Interviewer 3 5, 6
Interviewer 2 4 7, 8
Noise 5 9, 10

This is the information that should be entered in the "Number in TextGrid" field in the FAVE-extract web interface.

What are the FAVE-extract options?

screen shot of the FAVE-extract option fields

FAVE-extract options. Click on the image for a larger version.

screen shot of a pop-up window with explanations for the FAVE-extract options

Explanatory pop-up window for FAVE-extract options. Click on the image for a larger version.

FAVE-extract allows you to specify a number of options, which are explained in more detail in the pop-up boxes that appear when you click on or hover your mouse over the small red question marks next to the options in the FAVE-extract interface. You can either individually select the parameter settings you want in the upper portion of the "Options" section of the FAVE-extract interface, or upload your own configuration file.

If you are uncertain about which options to use, we recommend using the pre-selected default options.

An overview of all FAVE-extract options, with their allowed parameters and explanations, can be found here.

Output files

Plotnik files vs. tab-delimited text files

screen shot of a normalized Plotnik output file

FAVE-extract output file: Plotnik (.plt) file. Click on the image for a larger version.

screen shot of a tab-delimited text output file

FAVE-extract output file: Tab-delimited text (.txt) file. Click on the image for a larger version.

FAVE-extract produces two different formats of output files: Plotnik files and tab-delimited text files.

Plotnik files are comma-delimited plain text files designed to work with the Plotnik vowel analysis program. They consist of two header lines containing background information about the speaker and the number of tokens included in the file, and then a large body of individual lines, with one vowel measurement per line. At the end of the file, a blank line separates mean data for all vowel classes from the preceding individual token measurements.

Plotnik files use the file extension ".plt".

Tab-delimited text files can be easily read by a number of computer programs, such as Excel or statistics programs. The formant data is presented as a tab-delimited table, with one line per vowel measurement. The file extension for these kinds of files is ".txt", the same as for regular plain text files.

If the outputFormat option is set to "both", both Plotnik files and tab-delimited text files will be returned to the user.

More details on the structure and content of the FAVE-extract output files can be found here.

Normalized vs. unnormalized formant values

screen shot of a normalized Plotnik output file

Unnormalized Plotnik (.plt) file. Click on the image for a larger version.

screen shot of a tab-delimited text output file

Normalized Plotnik (.pll) file. Click on the image for a larger version.

In order to enable comparisons between vowel formant data for different speakers, FAVE-extract returns both normalized and unnormalized ("raw") formant data to its users.

For the normalization, FAVE-extract follows the Lobanov (1971) normalization procedure, which transforms a speaker's vowel space into z scores using the overall grand mean and standard deviation for the whole vowel system. These z scores are then re-scaled to Hertz values using the following settings:

mean standard deviation
F1 650 Hz 150 Hz
F2 1700 Hz 420 Hz

Normalized Plotnik files use the file extension ".pll" (instead of ".plt" for "regular" (unnormalized) Plotnik files).

The file names of normalized tab-delimited text files end in "_norm.txt".