Phonetic Exercises

Phonetic Exercises - Languages in Danger

Author: Maciej Karpiński
Keep on grooving

Rhythm of speech

Rhythm of speech is a complex phenomenon. We can hear it there but still we know that (besides reciting poetry) we just do not simply repeat the same patterns all the time. Speech deprived of its proper rhythm becomes incomprehensible. One of the factors contributing to our perception of rhythm is the difference in the duration and intensity of various segments that we are able to detect in utterances (e.g., of syllables).

Download the second sound file (MB_inka_pintsi) from:

Open the file in Praat in [View & Edit] window or in WaveSurfer, in [Speech Analysis] mode. Listen to the speaker for a while.

Then zoom in on a shorter portion of the signal, for example, from the 81st to the 82nd second of the recording. Listen to it closely.

Divide the file into syllables.

In Praat, you can do it by creating a TextGrid and placing markers in it (for details, see one of available tutorials – the main idea is that you create a new object (TextGrid) and then edit it together with the sound so that you can achieve synchronised transcriptions). You may face some difficulties but do not worry – just do your best and it will certainly suffice.
Before measuring anything, formulate a hypothesis on the possible proportion of syllable durations in this portion of speech: How much longer is the longest syllable than the shortest one? Measure and compare the durations of various syllables. What is the maximum and the minimum syllable length in the selected section of the narrative? What is the proportion of the shortest vs. the longest one?

[Although the syllable seems to be a relatively intuitive unit, the division into syllables is often ambiguous and difficult in practice. Here, you just give it a try – as you probably do not speak Cashinahua, segmentation may be even more difficult. But do not worry about your doubts now. In the suggested portion of the speech, the last syllable seems to be the longest, almost 500ms. The penultimate one seems to be very short but, as it starts with a plosion, the preceding period of silence should be counted as belonging to it which makes it twice as long, ca. 250ms. The antipenultimate (third from the end of a word) one is, in fact, shorter, with the duration of less than 100ms! The long one is five times as long as the short. And such a proportion is still not an extreme case. If you look in other portions of the signal, you will find even shorter and even longer syllables.]

Select a stretch of speech. Perhaps somewhat longer than the initial one.

Segment it into syllables (if you use Praat or WaveSurfer). Otherwise, just represent syllables as boxes on a sheet of paper. Now, listen to the utterance once again and mark the syllables that sound more prominent to you. Ask you friend(s) or family to do the same. How similar are your choices? Are you all completely coherent in your choices?

[The perception of prominence is a complex phenomenon and it is cued and biased by a number of more or less controlled factors. You can be surprised to find that a non-native speaker of your native language marks as prominent (in an utterance in your language) syllables which are not prominent according to your judgement! Of course, the same may happen to you when you mark prominences in Cashinahua. But being right or wrong in this respect is not crucial here! Just trust your ear and intuition.]

Now try to activate pitch and intensity view in Praat.

Find which parameters of syllables (durations, intensity, pitch) correlate with your choices from the earlier part of the exercise. Are the syllables you have marked as prominent longer than the others? Do they have higher energy levels? Or higher pitch values? Or maybe more pitch variation inside (e.g., rises, falls)?

[All of these factors as well as some others may contribute to the perception of prominence. You may find that some short or relatively quiet syllables are prominent only because of a significant pitch change that happens within their limits. On the other hand, some very long syllables can be perceived as not very prominent. And prominence should be judged against the context. Something is prominent in a given context but not necessarily prominent in another one.]

Try to tap with your finger the rhythm of the selected stretch of speech. Play the signal once or twice to memorize its rhythmic structure.

Open [Record mono sound] window from the [New] menu. Play the signal directly from the [Object] window of Praat and activate recording. Tap to the rhythm of speech. Save the recording to the list of Praat objects. Now you can listen to your own efforts. How precise were you? Ask you friend(s) or family to participate and do the same. Compare your efforts. Who was the closest to the original? Are the patterns you all produced mutually similar or not?

[Simultaneous playback and recording may cause some software or hardware problems but it should work on most computers. If not, use another computer or a playback device to play the sound.]