Development of Coarticulatory Patterns in Spontaneous Speech

By Dr. Jane Saalenberg

This study seeks to better understand natural speech processes by examining coarticulatory patterns in the spontaneously produced speech of adults and children. Coarticulation, the process by which any articulatory gesture affects adjacent articulatory gestures, can be anticipatory (as when knowledge of an upcoming gesture affects the realization of the gesture currently being executed) or perseverative (when an already initiated gesture carries over onto the articulatory realization of a following gesture).

We examine three acoustic measurementsin an attempt to distinguish productions of [s] in round vs. non-round vowel contexts, and we compare the results for adults vs. children. Experiment Corpora. The adult data in this study come from the Buckeye Corpus of Conversational Speech (Pitt et al., 2007), and the child data come from the Davis corpus of the CHILDES database (Davis et al., 2002; MacWhinney, 2000). Both corpora were phonetically transcribed by their respective developers, making it possible to identify all instances of [s] in the context of either a high rounded vowel ([u ʊ o]) or an analogous unrounded vowel ([i ɪ e ɛ]) using an automated computer script. The child data were then hand segmented by the first author and a research assistant1 . Table 1 shows the number of [s] tokens analyzed, broken down by corpus, direction of coarticulation (anticipatory vs. perseverative) and adjacent vowel type (round vs. nonround). For the child data, only tokens of [s] that occurred in identifiable words were used.

Procedure and analysis. We report three acoustic measures: high frequency centroid, amplitude ratio, and kurtosis. A description of each measurement is provided below. To calculate the acoustic measures, a 40 millisecond Hamming window was centered at three time points: 20%, 50%, and 80% of the fricative’s duration. (Fricatives shorter than 100 ms were excluded from the analysis.) A fourth window, centered at 20 ms into the adjacent vowel, was used to generate vowel spectra for comparison. The spectra were then averaged across all tokens and used to produce the plots in Figure 1. The left panel of Figure 1 shows the averaged fricative spectra for the anticipatory direction, based on the 80% duration time slice, juxtaposed with the adjacent vowel spectra. The right panel shows the perseverative spectra for the 20% duration time slice, also juxtaposed with the adjacent vowel spectra. We now describe the measurements used to characterize the differences in spectral shape.