Comparison of auto-contouring and hand-contouring of ultrasound images of the tongue surface.

Contours traced by trained phoneticians have been considered to be the most accurate way to identify the midsagittal tongue surface from ultrasound video frames. In this study, inter-measurer reliability was evaluated using measures that quantified both how closely human-placed contours approximated each other as well as how consistent measurers were in defining the start and end points of contours. High reliability across three measurers was found for all measures, consistent with treating contours placed by trained phoneticians as the 'gold standard.' However, due to the labour-intensive nature of hand-placing contours, automatic algorithms that detect the tongue surface are increasingly being used to extract tongue-surface data from ultrasound videos. Contours placed by six automatic algorithms (SLURP, EdgeTrak, EPCS, and three different configurations of the algorithm provided in Articulate Assistant Advanced) were compared to human-placed contours, with the same measures used to evaluate the consistency of the trained phoneticians. We found that contours defined by SLURP, EdgeTrak, and two of the AAA configurations closely matched the hand-placed contours along sections of the image where the algorithms and humans agreed that there was a discernible contour. All of the algorithms were much less reliable than humans in determining the anterior (tongue-tip) edge of tongue contours. Overall, the contours produced by SLURP, EdgeTrak, and AAA should be useable in a variety of clinical applications, subject to spot-checking. Additional practical considerations of these algorithms are also discussed.

Effects of Ultrasound Familiarization on Production and Perception of Nonnative Contrasts.

BACKGROUND/AIMS: We investigated the efficacy of ultrasound imaging of the tongue as a tool for familiarizing naïve learners with the production of a class of nonnative speech sounds: palatalized Russian consonants. METHODS: Two learner groups were familiarized, one with ultrasound and one with audio only. Learners performed pre- and postfamiliarization production and discrimination tasks. RESULTS: Ratings of productions of word-final palatalized consonants by learners from both groups improved after familiarization, as did discrimination of the palatalization contrast word-finally. There were no significant differences in the improvement between groups in either task. All learners were able to generalize to novel contexts in production and discrimination. The presence of palatalization interfered with discrimination of word-initial manner, and ultrasound learners were more successful in overcoming that interference. CONCLUSION: Ultrasound familiarization resulted in improvements in production and discrimination comparable to audio only. Ultrasound familiarization additionally helped learners overcome difficulties in manner discrimination introduced by palatalization. When familiarizing learners with a novel, nonnative class of sounds, a small set of stimuli in different contexts may be more beneficial than using a larger set in one context. Although untrained production can disrupt discrimination training, we found that production familiarization was not disruptive to discrimination or production.

Indexing head movement during speech production using optical markers.

Optical marker tracking integrated with electromagnetic articulometry was used to assess the movement extent of various points on (a) forehead skin and (b) points on a head-mounted apparatus, relative to a fixed point just above the upper incisors, and to compare the accuracy of the two different approaches to indexing head position during speech production. Both methods can provide a satisfactory index of head position. If skin-affixed markers are used, a minimum of 4 is recommended. Locations for optimal marker placement are identified.

Effects of phonological features on reading-aloud latencies: A cross-linguistic comparison.

Most psycholinguistic models of reading aloud and of speech production do not include linguistic representations more fine-grained than the phoneme, despite the fact that the available empirical evidence suggests that feature-level representations are activated during reading aloud and speech production. In a series of masked-priming experiments that employed the reading aloud task, we investigated effects of phonological features, such as voicing, place of articulation, and constriction location, on response latencies in English and Russian. We propose a hypothesis that predicts greater likelihood of obtaining feature-priming effects when the onsets of the prime and the target share more feature values than when they share fewer. We found that prime-target pairs whose onsets differed only in voicing (e.g., /p/-/b/) primed each other consistently in Russian, as has already been found in English. Response latencies for prime-target pairs whose onsets differed in place of articulation (e.g., /b/-/d/) patterned differently in English and Russian. Prime-target pairs whose onsets differed in constriction location only (e.g., /s/ and /ʂ/) did not yield a priming effect in Russian. We conclude that feature-priming effects are modulated not only by the phonological similarity between the onsets of primes and targets but also by the dynamics of feature activation and by the language-specific relationship between orthography and phonology. Our findings suggest that feature-level representations need to be included in models of reading aloud and of speech production if we are to move forward with theorizing in these research domains. (PsycInfo Database Record (c) 2022 APA, all rights reserved).

Perceiving while producing: Modeling the dynamics of phonological planning.

We offer a dynamical model of phonological planning that provides a formal instantiation of how the speech production and perception systems interact during online processing. The model is developed on the basis of evidence from an experimental task that requires concurrent use of both systems, the so-called response-distractor task in which speakers hear distractor syllables while they are preparing to produce required responses. The model formalizes how ongoing response planning is affected by perception and accounts for a range of results reported across previous studies. It does so by explicitly addressing the setting of parameter values in representations. The key unit of the model is that of the dynamic field, a distribution of activation over the range of values associated with each representational parameter. The setting of parameter values takes place by the attainment of a stable distribution of activation over the entire field, stable in the sense that it persists even after the response cue in the above experiments has been removed. This and other properties of representations that have been taken as axiomatic in previous work are derived by the dynamics of the proposed model.

Perceptuo-motor effects of response-distractor compatibility in speech: beyond phonemic identity.

Previous studies have found faster response times in a production task when a speaker perceives a distractor syllable that is identical to the syllable they are required to produce. No study has found such effects when a response and a distractor are not identical but share parameters below the level of the phoneme. Results from Experiment 1 show some evidence of a response-time effect of response-distractor voicing congruency. Experiment 2 showed a robust effect of articulator congruency: perceiving a distractor that has the same articulatory organ as that implicated in the planned motor response speeds up response times. These results necessitate a more direct and specific formulation of the perception-production link than warranted by previous experimental evidence. Implications for theories of speech production are also discussed.

Masked primes activate feature representations in reading aloud.

Theories of reading aloud are silent about the role of subphonemic/subsegmental representations in translating print to sound. However, there is empirical evidence suggesting that feature representations are activated in speech production and visual word recognition. In the present study, we sought to determine whether masked primes activate feature representations in reading aloud using a variation of the masked onset priming effect (MOPE). We found that target nonwords (e.g., BAF) were read aloud faster when preceded by masked nonword primes that shared their initial phoneme with the target (e.g., bez), or primes whose initial phoneme shared all features except voicing with the first phoneme of the target (e.g., piz), compared with unrelated primes (e.g., suz). We obtained the same result in 2 experiments that used different participants and prime durations (around 60 ms in Experiment 1 and 50 ms in Experiment 2). The significant masked feature priming effect that was observed in both experiments converges with the empirical evidence in the speech production and visual word recognition domains indicating a functional role for features in reading aloud. Our findings motivate the further development of current theories of reading aloud and have important implications for extant theories of speech production.