MEASURING PRE-SPEECH ARTICULATION
Abstract: What do speakers do when they start to talk? This thesis concentrates on the articulatory aspects of this problem, and offers methodological and experimental results on tongue movement, captured using Ultrasound Tongue Imaging (UTI). Speech initiation occurs at the start of every utterance. An understanding of the timing relationship between articulatory initiation (which occurs first) and acoustic initiation (that is, the start of audible speech) has implications for speech production theories, the methodological design and interpretation of speech production experiments, and clinical studies of speech production. Two novel automated techniques for detecting articulatory onsets in UTI data were developed based on Euclidean distance. The methods are verified against manually annotated data. The latter technique is based on a novel way of identifying the region of the tongue that is first to initiate movement. Data from three speech production experiments are analysed in this thesis. The first experiment is picture naming recorded with UTI and is used to explore behavioural variation at the beginning of an utterance, and to test and develop analysis tools for articulatory data. The second experiment also uses UTI recordings, but it is specifically designed to exclude any pre-speech movements of the articulators which are not directly related to the linguistic content of the utterance itself (that is, which are not expected to be present in every full repetition of the utterance), in order to study undisturbed speech initiation. The materials systematically varied the phonetic onsets of the monosyllabic target words, and the vowel nucleus. They also provided an acoustic measure of the duration of the syllable rhyme. Statistical models analysed the timing relationships of articulatory onset, and acoustic durations of the sound segments, and the acoustic duration of the rhyme. Finally, to test a discrepancy between the results of the second UTI experiment and findings in the literature, based on data recorded with Electromagnetic Articulography (EMA), a third experiment measured a single speaker using both methods and matched materials. Using the global Pixel Difference and Scanline-based Pixel Difference analysis methods developed and verified in the first half of the thesis, the main experimental findings were as follows. First, pre-utterance silent articulation is timed in inverse correlation with the acoustic duration of the onset consonant and in positive correlation with the acoustic rhyme of the first word. Because of the latter correlation, it should be considered part of the first word. Second, comparison of UTI and EMA failed to replicate the discrepancy. Instead, EMA was found to produce longer reaction times independent of utterance type.Keywords: Speech initiation, pre-speech articulation, delayed naming, ultrasound tongue imaging, electromagnetic articulography, automated methods.