LAB 1: Praat and English consonants

What to turn in, and when

For Part I: nothing.

For Part II: send your TextGrid as an email attachment to Joshua Tauberer. If you use your own numbering system for allophones, indicate in the body of the email what that numbering system is. Please do this before the lab next week.

Part I: Using Praat

There are a lot of online Praat tutorials (try google: Praat tutorial), many of them are very good. We recommend the one written by Jean-Philippe Goldman (available at http://www.unige.ch/lettres/linge/ppp/praat_tutorial.pdf). You may start with any tutorial you found good, as long as you can learn how to do the following:

• Record and load a sound;
• Open an edit window, select a part of the sound in the edit window;
• Create a TextGrid with proper tier(s), and do segmentation and labeling.

Here are some suggestions on using a head-mounted microphone to make decent-quality recordings directly into your laptop:

• Choose a good microphone. A very important specification of a microphone is its frequency response. An ideal microphone (not exist yet) would have a flat response for all frequencies. Generally speaking, a frequency response that is flat (within 3dB) from 100 to 10000Hz is adequate for speech. You can also choose the new emerging microphones with a single USB plug-and-play connection (They are slightly more expensive, but have higher quality), for example, Samson C01U and Logitech Premium USB Headset 30.
• Set proper recording levels (for windows, go to Control Panel > Sound and Audio Devices > Audio > Sound recording > Volume ...). It's important to set the recording level high to improve SNR (Signal to Noise Ratio), but you don't want to set levels so high that the microphone or sound card is overloaded. SNR is the ratio of the signal level to the noise level. In general, an SNR greater than 50dB is adequate for the purpose of phonetic studies. You can estimate SNR by following these steps:
  • 1. Load your recording (or make a new recording) into Praat;
  • 2. Edit the sound object, go to the edit widow, make sure that View > Sound autoscaling is checked on (see illustration);
  • 3. Go to the left bottom corner of the edit window, click on all. There are two numbers shown on the left side of the sound wave window. The top one is positive and the bottom one is negative. These are the maximum air pressures of the sound (only if Sound autoscaling is set on). Record the greater absolute value of these numbers (see illustration). This is the maximum air pressure of the signal, P_signal, measured in Pa (well, not exactly, please refer to Praat documentation for more information);
  • 4. Select a silent portion (no speech signal in it) of the recording, click on sel to show the selected portion only. Again, record the greater absolute value of the two numbers shown on the left side of the sound wave window. This is the maximum air pressure of the noise, P_noise;
  • 5. Calculate SNR (in dB) of your recording: SNR = 20log(P_signal/P_noise), where log is 10-based;
  • 6. Is your SNR greater than 50 dB? If not, try to set your recording level higher (be careful not to overload), or move to a quieter place, and record again.
• For more information about microphone and recording, please refer to these excellent online documents: Microphone, Recording.

Practice:

2. Download file try1.wav , open Praat, load try1.wav into Praat by clicking on Read > Read from file ... , and selecting the file "try1.wav" you just downloaded. A highlighted sound object called "try1" will appear in the object window (looks like this).

3. Click on Edit, an edit window will pop up. If no settings of Praat have been changed, the upper part of the window will display the waveform and the pulses of the sound, and the lower part will display the spectrogram, formants, pitch and intensity (looks something like this -- depending on your default settings, details may differ).

4. Click Pulses on the top bar of the edit window, then uncheck Show pulses (or check it if it's unchecked to start with.) The pulses shown in the upper window will now disappear (or appear). Repeat the same procedure for Spectrum (check/unckeck Show spectrogram), Formant, Pitch, and Intensity. Fiddle with the checks until your edit window looks (more or less) like this.

5. Go back to the object window, click on Annotate, then choose To TextGrid... A small window will pop up. Replace "Mary John bell" on the top line with "word", and delete "bell" from the bottom line, and then click OK. By doing this, you'll create a TextGrid which has only one interval tier.

6. In the object window, a TextGrid object, also called "try1", will show up and be highlighted. Now press and hold the 'Ctrl' key - the Cmd ('apple') key on a Mac -- and click the sound object. You will see that both the sound and the TextGrid objects are now selected/highlighted. Click Edit on the right side of the object window, and you'll see a new edit window like this.

7. Experiment with changes in the amount of speech shown (e.g. via the "all, in, out, sel" buttons at the left-bottom corner of the edit window, or via View>>Zoom to selection) and with playing segments of the audio file (e.g. using the tab key, or clicking on the horizontal bars at the bottom of the edit window).

8. Now click on the beginning of the first word, and you'll see a vertical cursor shown in the TextGrid window. There is a small circle on the top of the curosr. By clicking on the circle, you'll add one point of an interval in the TextGrid. Now move the mouse to the end of the first word, click, and click the small circle, the second point of the interval is then added. Click on any point between the two cursors, you will see the entire interval becomes yellow. Write the word on the plane of the interval. If you want to move a cursor, just click on it and drag it; to remove a cursor, click on it, then select Boundary > remove from the top bar, or press Alt+Backspace. Congratulations! Now you know how to segment and label using Praat. Be sure to use the "all, in, out, sel" buttons at the left-bottom corner of the edit window to show the entire file, zoom in, zoom out, and show the selected portion. You will need to listen to different portions of sound when you do segmentation. You can do so by clicking on the horizontal bars shown at the bottom of the edit window.

9. Segment and label all the words in try1.wav. After you are done, you should get a window like this.

10. Estimate the SNR of try1.wav. It should be about 60dB.

Part II: English consonants

Download this speech paragraph , which is from the Boston University Radio News corpus. Using the Praat textgrid feature, enter an orthographic transcription (i.e. in ordinary English spelling). Now classify the instances of the (dictionary) phoneme /t/ found in this passage, and enter them into a second tier on the textgrid.

[You probably want to create the textgrid with two appropriately-labelled tiers to start with. But you could also experiment with adding a new tier to an existing textgrid.]

You can use IPA symbols to do this; but a more convenient (and more revealing) way might be to make a separate list of allophonic categories, e.g.

1. Prevocalic aspirated stop
2. Intervocalic voiced flap
3. Glottalized syllable-final stop
4. ...

and then use the numbers as category labels.

Alternatively, you could simply number the /t/-regions from 1 to 20 in the text tier, and provide a separate numbered key with a classification and description of each one. This third approach has the benefit of allowing you to change your mind as your research progresses, without having to go back and change all the textgrid labels (which is more trouble than editing a text file is).

Extra credit: do the same sort of thing for one or two additional consonant phonemes.