The perception and production of rhythmic sequences
5.2 Experiment 2a Production 1 Method
Sequences of rhythmic speech were collected from eight subjects, three male and five female. Subjects were instructed to take a breath and then to repeat the sequence "one..two..one..two..etc" until they need to take another breath. As the subjects were instructed in the perception experiments (see Chapter 4) they were told to time the words ‘evenly’, that is at intervals which feel regular. Their rate of speaking was to be whatever speed felt most comfortable. They were instructed to speak faster or slower if the experimenter felt they were ‘rushing’ their words, or speaking too slowly. Speakers who spoke too fast would produce speech which could not be analyzed in terms of onset to onset intervals. Speakers who spoke too slowly would not produce a rhythmic sequence at all. Several training trials were given with the experimenter present providing feedback. The feedback provided was restricted to help with speaking rate, guidance as to the rhythm to be acheived, and encouragement. These were recorded and the subject could hear the tape back if they wanted to. When both the subject and the experimenter were happy with the performance, the subject produced some more sequences. The tapes were then listened to by two raters (one the experimenter), who chose the perceptually regular sections of the utterances - that is, selecting out the offset sections of the sequences, where the speakers slowed down, and the sections at the onset of a sequence where speakers would normally take a couple of utterances before producing rhythmic speech. The sections were then digitized at 20kHz and the onsets annotated by hand from an oscillogram display which afforded magnification of sections of the display. Initial deviations of the time / amplitude plot were used to denote the physical onset. This measurement technique was repeated to improve consistency and accuracy.
5.2.2 Analysis
From the spoken sequences "one two one two", it was the one-two and two- one onset to onset intervals that were of interest. The relative sizes of these two intervals was a measure of the physical anisochronies that results in perceptual isochrony for each of the speakers. Therefore, for each speakers, these onset to onset intervals had to be measured and the means calculated.
5.2.3 Results
The results of one speaker (SKS) had to be discarded. This speaker failed to produce a perceptually isochronous sequence. Although they tended to be confident at the time of production, they would agree on a later hearing that the sequences were irregular. After several attempts their sequences were not analyzed any further, although their speech was used in a later rhythm setting experiment. Speech from seven speakers (three male, four female) was therefore analyzed. The means and S.D.’s for these speakers are shown in Table 5.1 below.
interval Types
Speaker One - Two intervals (ms) Two - One intervals (ms)
SHS 703.10 (45.30) 700.57 (30.32) PH 722.15 (64.9) 630.57 (35.55) LCE 681.85 (23.50) 682.38 (31.58) SM 504.10 (41.70) 587.40 (28.44) DG 613.24 (41.99) 660.04 (30.59) WC 659.34 (36.23) 702.23 (39.93) SR 742.66 (35.06) 800.53 (19.68)
Table 5.1 of means and S.D’s of produced (one - two) and (two - one) intervals
Chapter 5 - Experiment 2
Table 5.1 shows that the speakers are producing perceptually even sequences with inter-onset intervals that vary across the two interval types and the speakers. Speakers LCE and SHS show no clear difference between the (one- two) and (two-one) mean intervals. Speakers DG WC SM and SR show larger (two-one) mean intervals, than (one-two). Speakers PH shows larger (one-two) mean intervals. The significance of these differences was tested statistically, using independent t-tests. Each speaker was examined separately, since all the speakers spoke at different tempos, and therefore direct comparison across speakers was not possible.
These produced intervals are plotted against speaker for each word type in Figure 5.1. Independent t-tests were applied to each set of intervals, to test the hypothesis that for each subject, the (one-two) and (two-one) intervals were equal in size. The results are shown in Table 5.2.
m a in In ta f'O n ta t In ta rv a l (m a) Figure 5.1 900 800 700 600 500 400 ■ (out • tae) ■ (two • ofio) tpaaker
Figure 5.1 Mean (one-two) and (two-one) intervals, for each speaker in production
Speaker T value p value degrees of freedom SHS t = 0.17 p = 0.86 df = 22 PH t = -4.56 p - 0.0002 # # 2 0 LCE t = 0.05 p = 0.96 df = 22 SM t = -6.53 p #0.0000 df = 26 DG t = -3.93 p = 0.0004 df = 32 WC t = -3.33 p = 0.0022 df = 32 SR t = -5.76 p = 0.0000 df = 23
Table 5.2 independent t-test results comparing the mean (one-two) interval with the mean (two-one) intervals for each subject. Cases where the difference between interval types are significant are shaded.
Thus for the subjects PH SM SR WC and DG the differences between the (one - two) and (two - one) intervals are significant. As mentioned earlier, these differences are net all of the same magnitude or direction. While LCE and SHS show no difference between their intervals, subjects PH and SM show large, and opposite effects: while SR WC and DG show similar, smaller variation, all in the same direction.
The absolute sizes of the intervals are not considered further, since each speaker was producing sequences at different rates. Rather, this pattern of interval differences will be compared to those set in the perception part of the experiment. It is sufficient here to note that:
a) There are differences in several cases in the relative sizes of the (one-two) and (two-one) intervals.
Chapter 5 - Experiment 2