This Vowel Logging System records F1, F2, and F3 values for nuclei and glides of vowels, as well as vowel duration, and duration of nucleus to glide transition. It also creates a labeled Praat text grid for the word, vowel, and measurement points.
This vowel logging system consists of two praat scripts
Open a sound file in Praat. Select the "Annotate" dropdown menu, then choose "To TextGrid..." What you choose to name the tiers is immaterial, but for the sake of readabilty, I'd suggest "word vowel points". In the box asking which tiers are point tiers, enter the name of the third tier (if you're following my advice, that'd be "points").
Open labeler.praat in the Praat script editor, or in any other plain text editor. The first few lines of the script (after the GNU license) will look like this
Next, open logger.praat. The first few lines of the script (again, after the GNU license information) will look like
Select both the sound object and the text grid in the Praat objects window, and click Edit. You should now be in the Sound Editor window. From the the 'Query' dropdown menu, select 'Log Settings...' The log settings window should open up. In the field labeled 'Log script 3' give the full path to the labeler.praat script. In the field labeled 'Log script 4' give the full path to the logger.praat script.
To start logging a vowel, select the full duration of the vowel in the sound editor. The beginning of the selection should be at vowel onset, and the end of the selection should be at vowel offset. Then hit the shortcut command for labeler.praat. If you set up the Log Settigs as described above, this will be the shortcut for Log Script 3. The actual keystrokes vary by system, but they are labeled next to each log script in the Query menu.
A dialog box will open. It will display the name of the text grid for sanity checking, and will have two dialog boxes for the word and vowel. Enter the word being logged in the Word field, and whatever label you are using for the vowels in the Vowel field and hit Enter.
labeler.praat will create labeled intervals on the word and vowel tier from the information provided. It will also concatenate the number of the measurement to the end of the word label. If you have logged 5 vowels already, and now are now logging "spider," the label on the Word tier will be "spider6". This should be useful for not confusing multiple tokens of the same word.
The last thing labeler.praat will do is create a new selection within the vowel, with the beginning at the F1 maximum, and the end at the F2 maximum.
If you want to measure both a nucleus and glide target for the vowel you're logging, make a selection with the beginning at the vowel nucleus, and the end at the vowel glide. The labeler.praat script will have created a selection beginning at max F1 and ending at max F2, which are pretty good measurement points for some diphthongs, but of course very inappropriate for others. If you want to just measure a nucleus, just place the cursor at the measurement point without making a selection. I would suggest making a point selection if there really is no glide whatsoever. Measuring a glide requires almost no additional work, and this way, you will be able to distinguish between vowels with glides and true monophthongs in your data.
After making the appropriate selection, hit the keyboard shortcut for Log Script 4. The logger dialog box will open. This is here strictly for sanity checking. It will display the text grid name, the output file name, and the nucleus and glide window information. By default, the logger.praat script will take the mean formant values from a window 5 ms before and after the selected nucleus point, and 2ms before and after the glide point. If you would prefer just to take point measurments, without averaging over a window, I would suggest openning the logger.praat script and changing 0.005 to 0, for the NucleusWindow variable, and likewise for the GlideWindow variable. If everything looks good in the logger dialog box, just hit enter.
The logger script will write the logged values to he output script, and create intervals on the point tier, appropriately labled 'nuc' and 'glide'.
The output of the logger.praat script is a tab delimited file with the following elements (listed left to right).
As stated above, the nucleus formant data is, by default, the mean value over a 10ms window centered around the nucleus time, and the glide formant data is an average over a 4ms window centered around the glide time.
The glide time, F1, F2 and F3 is printed for all vowels, even those for which just a nucleus was measured. For these monophthongs, the glide data will be the same as the nucleus data. Two ways to pick out the monophthongs from the rest will be that their Nucleus to Glide Duration will be 0, and their Has Glide? value will be NoGlide.
With any luck, the process of automated formant analysis will be refined in the near future, especially given the recent successes of Keelan Evanini's dissertation (link). In the meantime, however, I hope this logging system will be useful for improving the scalability and recoverability of hand measurements.
It seems to me that lot of hand measurement studies, like the Atlas of North American English, or Dinkin's Herculean study of New York State (one man, 57,664 hand measurments), lose as much data as they collect. Every observation consists of a single point measurement with values for F1, F2 and F3, along with a time stamp, and perhaps an additional point measurement for the glide.
Usually no data is recorded for the duration of the vowel, which is clearly a dimension of interest for the study of vowel changes (see Baranowski and Labov, 2006). Additionally, a clear record of the observation is not recorded which will allow for a rapid restudy. For example, a restudy of the ANAE data focusing on the role formant dynamics on merger (see Scanlon 2009), perhaps by utilizing Smoothing Spline ANOVAs (see Baker 2006, Nycz and De Decker 2006, Koops 2009 and my write-up here [PDF]), would be virtually impossible without redoing the study.
However, these drawbacks to a manual study are strictly due to the collection tools used. It is completely possible to rapidly record duration data along with nuclear and glide formant data, while at the same time leaving yourself a trail of breadcrumbs should you want to restudy. According to my own general principles to data collection, when working with observational data it is crucial to answer the following question: "How can I preserve a record of this observation in such a way that I can quickly return to it and gather more data on it if necessary?" This vowel logging system is my own approach to answering this question.
The time-per-observation to data-per-observation ratio is rather good. I haven't noticed it to be significantly slower than the time it takes to make a single point measurement, and in the end you have data on more dimensions, and a clear trail of crumbs to follow back to the original observation so you can collect more data from it if needed.