Linguistics 580
General Phonetics
To become familiar with articulatory movement data; to learn to identify gestures of the labial, coronal, and dorsal constrictors, in traces showing movements of articulators over time.
You will be analyzing recordings of the following sentence:
"Your good pants look great, however, your ripped pants look like a cheap version of K-mart special."
as produced by a speaker whose articulations were being simultaneously tracked using the X-ray microbeam system. You will be looking at synchronized displays of four traces over time: the audio (microphone) signal, and the vertical position of pellets placed on the lower lip, tongue tip and tongue dorsum. When the position of one of these pellets goes from relatively low to relatively high, this means that a constriction is being formed by the relevant constrictor. The formation of this constriction over time is what is referred to as a gesture. This exercise involves identifying, in the
appropriate pellet traces, the periods of movement corresponding to
the constriction gestures for each of the stop consonants and
fricative consonants in this sentence .
The sentence can be transcribed in a broad phonetic transcription as follows:
The stop (including nasal) and fricative consonants are shown in
bold in the transcription. For each of these consonants you are to
locate the constriction gestures in the movement of the relevant
pellets--the Lower Lip pellet for labial gestures, the Tongue Tip for
coronal gestures, and the Tongue Dorsum for dorsal gestures.
Praat (Dutch for "speech") is a system for recording, analyzing, and synthesizing speech, developed at the University of Amsterdam. We will be using it for several exercises. Praat is available o in the Phonetics Lab, and the instructions below and in the future will describe its use on those machines. You download a copy of Praat for your own computer by visiting http://www.praat.org, and following the instructions there. Both Mac and Windows versions are available. Note that the instructions are designed to work with Mac computers. There may be some differences in how your own copy works, particularly in regard to sound recording, as this will depend on your hardware and software configuration.
1. Download the files we will be analyzing here (one for speaker A1, one for speaker D2).
2. Launch Praat. Go to the Read menu of Praat, and select Read from file... Navigate to you folder into whixh you downloaded the files and select the file D2.Collection,
3. The Praat Objects window will show the various sound signals and other kinds of data available in the D2 collection (See Figure 1) . If you select one (for example, Sound D2_audio is selected in Figure 1), then various operations that can be performed to/with this object are shown as buttons in the right-hand column.
4. The object Sound D2_audio contains the signal recorded by the microphone when the speaker produced the K-mart sentence. We can examine it and can listen to it or parts of it. The other objects contain the X-ray data of the subject's vocal tract and annotations of that data. For example, the object Sound D2_lip contains the X-ray data showing how the vertical position of the pellet on the lower lip moves over time, and TextGrid D2_lip contains an annotation of the lip signal, indicating where the constriction gestures are located. We will focus first on looking at the audio signal and listening to it.
Audio Signal
5. To examine and play an audio signal, click on the desired sound object (e.g., D2_audio), then click on the Edit button in the right-hand column. A new window will appear that will look something like that in Figure 2. The display shows the signal with various "play bars" below it. If you see additional displays (spectrograms, pitch, etc.) go to the window's View menu, select Show analyses... and uncheck all of the entries.
Figure 2
6. Playing the sound. Clicking on the play bars below the acoustic waveform will play the signal or parts of it. The top play bar plays intervals of the waveform displayed in the current window. Note the red cursor line in the audio waveform. It can be moved by clicking anywhere in the waveform display. The cursor divides the signal into intervals, and the play bar is also divided. Clicking on the left side of the bar plays the audio signal to the left of the cursor, clicking on the right side of the bar plays the signal to the right of the cursor. Note also the numbers on the play bars. These represent the duration of the respective intervals (in seconds). Clicking on the middle play bar plays the entire signal in the current window. Clicking on the bottom play bar plays the entire sound object (which can be larger than the current window, if zooming has taken place).
7. Selecting an interval. To select an interval of the sound waveform, click somewhere in the display, hold down the mouse and drag. A segment of the wave display will be highlighted in blue. For example, select the section of the signal that includes the first three big "humps," as in Figure 3. Note that now the top play bar is divided into three intervals. Hitting the Tab key will play the selected interval. The selected interval should correspond to "Your good pants"
Figure 3
8. Zooming. The time scale of the display can be changed to provide a more detailed picture of the waveform. By clicking on the sel button at the lower left of the window, the scale is changed so that a selected interval now fills the entire window. For example, clicking the sel button of Figure 3, produces as display like Figure 4. To return to the full wave display, click on the all button. The in and out buttons zoom to a finer or coarser temporal display, respectively.
Figure 4
Movement signals and annotations
We can examine the vertical movements of the lower lip, tongue tip, and tongue dorsum pellets using the other signals in the list.
9. The first stop or fricative in the sentence is the /g/ in "good", so let's open the dorsum signal first. Select both Sound D2_dorsum and TextGrid D2_dorsum in the Object list (by holding down the shift key as you select). Then click on the Edit button in the right-hand column. A new window will open that looks like that in Figure 5. The top panel shows the vertical position of the tongue dorsum over the course of the sentence, the bottom panel shows the intervals of that signal that correspond to particular gestures (Only one is marked now. You are going to supply the rest).
Figure 5
10. Even though Praat calls the dorsum signal a sound object, it is not really a sound.. If you click on the play bars, you won't hear anything (try it). This is because it doesn't contain a record of sound vibrations, but rather a record of articulator movements.
11. The dorsum signal can be synchronized with the audio sound signal by checking the Group box in the lower right-hand corner of the dorsum window. You should then see the dorsum display change, as you will see the segment corresponding to that shown in the audio signal, as in Figure 6.
12. Note that there is "hump" in the dorsum curve around the time labeled with the letter "g." You can see exactly what part of the curve is annotated as "g" by clicking on box containing the "g." Now just that segment of the dorsum signal and the corresponding audio signal are selected and highlighted in blue. The annotated segment begins at the point at which the tongue dorsum begins to raise to form the velar stop and ends at the point at which the dorsum begins to lower so as to release the stop. This interval was selected to represent the velar stop gesture. If you click on the play bar of the audio signal corresponding to this interval, you will hear something not entirely clear. That is because our perception of gestures depends a lot on their release, as well as their formation. Here we are annotating only the closure portion. To hear the /g/ clearly, we need to select a larger chunk that includes some of the release. Vocal tract closure corresponds to the region of lowered amplitude in the audio signal. To hear the release, you have to extend the selection to the right, where the audio amplitude becomes large again. You will see this corresponds to the downslope of the "g" hump in the dorsum signal.
13. The next stop or fricative in the sentence is the /d/ in "good." To see the tongue tip gesture involved, select the D2_tip (Sound and TextGrid) and click on Edit. When the window opens, check the Group box on the lower right. Now note the region labeled "d," corresponding to the alveolar stop gesture. Here you the raising and lowering of the tongue tip for last consonant in "good.". You can confirm this by listening to the audio signal during the corresponding interval.
14. The next consonant is the initial /p/ in "pants." Select the D2_lip and note the interval labelled "p"
1. Find the approximate section of the audio signal that contains the next stop or fricative by zooming out a bit, scrolling, and/or selecting portions and listening. When you have located it, select a small region (about .5 seconds) that includes the relevant consonant.
2. Examine the signal (lip, tip, or dorsum) appropriate for the consonant, and find the hump (raising then lowering) corresponding the relevant gesture. Position the cursor at the onset of the raising movement--where the curve first starts to rise. A small circle should appear in the TextGrid panel, as shown in Figure 7 (for the /t/ in "great." ) Click in the circle, and a red boundary marker appears at that instant in time.
Figure 7
3. Find the point in time at which the release of the constriction begins--when the curve first begins to move abrptly down. Position the cursor and click in the circle to produce another boundary marker.
4. Click in the TextGrid between the two boundary markers. The selected interval will be highlighted in yellow. Type the phonetic symbol for the consonant and it will appear between the markers. For the consonants that require special symbols, use the following substitutions:
5. Continue annotating each consonant in sequence. Important. Note that sequences of adjacent consonants that involve gestures of the same constrictor may not show separate humps ([nts] in "pants" is an example). In such a case, select a single interval from the onset of movement for the first consonant to the end of an observable "plateau" (even if the plateau is somewhat jagged, reflecting differences between the consonants). Label the interval with the whole sequence of consonant symbols (e.g. "nts"). This interval is already labelled for you.
6. Note that there will be extra humps in each of the tracks that correspond to vowels and to consonants like /r/ and /l/. /r/, in particular, shows a very large movement of the tongue tip (as well as lip movements). You can ignore these.
7. Periodically save the signals to a file on the hard disk, so you don't lose your work. Select all the objects in the object list and go to the Write menu and select Write to binary file... Save the file as D2_<your name>.Collection.
(8) If you think you have made a mistake and you want to get rid of a labeled interval, click on the boundaries (one at a time) and select Remove from the Boundary menu of the signal window.
(9) When you have annotated all the gestures, select all the objects in the Object list and save again as above in (7).
(10) Repeat the exercise for the other utterance in the folder (A1). When both are done, email the the files to me as attachments,