1.1 What are phonetics analyses anyway?

Praat assists us in carrying out phonetic analyses of speech sounds, but just what are phonetic analyses anyway? Roughly stated, they are comparisons of acoustic measurements between two or more groups of speech sounds, and Praat helps us deal with the practicalities of grouping speech sounds of interest (e.g. word-initial vs. word-medial vowels, interrogative vs. declarative sentences), and extracting acoustic measurements of interest (e.g. pitch, intensity, formant structure) for comparison.

1.2 Goal: A short report of your vowels

The goal of the workshop will be to produce a short report with 4 sections (described below in the side notes). Don’t worry if you don’t understand all of the vocabulary in the report text—you’ll be familiar with them by the end of the workshop!

In this report, we compare first and second formant averages of five vowels reported for Australian-English speakers in Cox (⊕2006Cox, Felicity. 2006. “The Acoustic Characteristics of/hVd/Vowels in the Speech of Some Australian Teenagers.” Australian Journal of Linguistics 26 (2): 147–79.) with averages collected from a participant NS, a 28-year-old male Burmese-English bilingual.¹⊕¹Describe the aim of your analysis and participant information. We recorded five words in the /hVd/ context (heed, hard, horde, who’d, hair’d) using the built-in microphone on a 2016 Macbook Pro in a relatively quiet meeting room at the Australian National University.²⊕²Describe items recorded and recording environment.

The vowels were annotated in the Praat Editor Window with only the spectrogram visible (pitch, intensity, and formant contours were turned off), and using the default spectrogram settings (View range: 0-5000 Hz, Window length: 0.005, Dynamic range: 70 dB).³⊕³Describe annotation environment and annotation criteria. As illustrated in the figure below, the starts of vowels were placed at the beginning of low frequency components (indicating the end of the voiceless fricative /h/), and the ends of vowels were placed at the termination of high frequency components (signalling the start of the stop closure /d/).

The calculated formant averages were located on a F1-F2 plane along with average values for AusE male and female speakers from Cox (⊕2006Cox, Felicity. 2006. “The Acoustic Characteristics of/hVd/Vowels in the Speech of Some Australian Teenagers.” Australian Journal of Linguistics 26 (2): 147–79.). As observed in the figure below, the acoustic data indicated that NS’s heed vowel was produced in a location effectively identical to the AusE Male average, while their who’d was relatively more backed than both the AusE female and male averages.⁴⊕⁴Produce vowel plot, and make some observations.

2.1 Vocal tract configuration and formant variation

All the theory you need for this workshop can be found in an excerpt from Joe Wolfe’s excellent video “Resonances & Formants” (1m17s to 2m32s):

I provide here a transcript of the excerpt (emphasis mine):

Here’s a simplified demonstration, I have a source with many different frequency components present, and a flexible tube. If I change the geometry of the tube, I change the frequencies of its resonances, and so change the formants.

If I open my mouth more, the tract is a bit more like a cone than a cylinder, so the first resonance rises; […and] conversely [falls], if I close it. So the mouth opening largely determines the first resonance and the formant.

Whether the tongue constriction is forward or back strongly influences the second resonance.

The vowels are traditionally located on a plot of mouth opening versus tongue position, which is almost the same as a plot of first formant versus second, with the axes reversed.

2.2 What is a spectrogram anyway?

Before we define what a spectrogram is, let’s have a look at the Swiss Alps! The figure below shows, separately, the height of various mountain peaks (e.g. Eigner: 3970 m) and their geographical location on Google Maps.⊕Sources: MySwitzerland.com Grindelwald Skimap and Google Maps on Eiger, Switzerland

One way that we can incorporate altitude information into a map is to use a color gradient to indicate the metres above sea level. For example, we can see a small patch of white in the circle below, indicating the group of mountain peaks well above 3000 m.⊕Map source: Wikimedia Commons

Now, let’s start by looking at two spectral slices, graphs indicating how much sound energy was detected (y-axis) for various frequencies (x-axis) at some given point in time:

Notice above how Slice B has a bigger bump at the start, compared to Slice A? In other words, there’s more acoustic energy at the lower frequencies in Slice B than in Slice A. We see this in the spectrogram as the dark band at the bottom of Slice B, and the relative lightness at the bottom of Slice A. So, just as a terrain map indicates how altitude varies by latitude and longitude, a spectrogram indicates how sound energy varies by frequency and time.⊕

3.1 Recording your vowels

1. Write down the words to be recorded (it’s much easier!), and do feel free to swap the order:
   • Heed
   • Hair’d
   • Hard
   • Horde
   • Who’d
2. Open up Praat, and from the top left menu item ‘New’, click Record mono Sound... which will open the SoundRecorder in a separate window.
3. Click Record to begin recording your vowels and Stop to finish
   • Pause for a bit between each word, and also give a second or so before and after your word list in order to make sure you capture all the words.
   • Press Play to playback the most recent recording, or Record again to overwrite it.
4. When you’re happy with the recording, give the sound file a name (e.g. my_vowels) in the Name field in the bottom right corner of the SoundRecorder window, and then press Save to list & Close. You will see your recording (e.g. ‘1. Sound my_vowels’) in the list of Objects.

3.2 Annotating your vowels

1. Select your Sound in the Objects list (e.g. 1. Sound my_vowels), or if you had pre-recorded your vowels before the workshop, click Open then Read from file....
2. Click the Annotate - button then To TextGrid..., and enter the values below:
3. In Objects list, select both your TextGrid and Sound files (Command + Click for MacOS, Control + Click for Windows), and then click View & Edit
4. Now we’re ready to annotate! We’re going to get you used to using shortcuts in Praat (Command for MacOS, Ctrl for Windows):
   • Cmd + N: Zoom to selection
   • Tab: Play selection
   • Cmd + 1: Add interval on tier 1
   • Cmd + A: Show all

⊕Click the animated image to view the video which you can play/pause to follow it at your own pace.

3.3 Data extraction

We’re going to do this step twice. Once manually, and then once semi-automatically with Google Sheets.⁵⊕⁵If you’re really keen and are familiar with R, you can also do this step in R! By the end of this stage, however you do it, you should have a table as the one shown below. Go to the plot interface to make an F1-F2 plot with these values.⊕ Note there are many ways of measuring formant values other than averaging over an interval, but for introductory purposes we’ll stick with the average for this tutorial.

Vowel	F1_average	F2_average
heed	313	2332
hard	737	1073
horde	367	766
who’d	306	1249
hair’d	397	1913

3.4 Plotting your vowels