Sarah M. Hamilton

Acquiring rhoticity across languages: An ultrasound study of differentiating tongue movements

ABSTRACT Rhotics in many languages develop late and show clinically significant misarticulations. The English approximant and Spanish trill rhotics exhibit both a primary constriction along the palate and a secondary constriction in the pharynx. We speculate that the secondary constriction might be a cross-linguistic characteristic of rhotics and thus potentially a factor in delayed articulatory development/misarticulations. We describe an exploratory study analyzing rhotic tongue configurations in ultrasound videos from a small sample of native adult speakers of English, Malayalam, French, Persian and Spanish. Our findings confirm that rhotic sounds most subject to late development also involve tongue root movement towards a pharyngeal constriction, but this conclusion must remain tentative without further research. In the meantime, clinical strategies that include attention to primary and secondary constrictions should be explored for remediation of rhotic misarticulations across languages.

An acoustic threshold for third formant in American English /r/

It is well known that a low F3 is the most salient acoustic feature of American English /r/, and that the degree of F3 lowering is correlated with the degree to which /r/ is perceptually acceptable to native listeners as a “good” vs. “misarticulated” /r/. Identifying the point at which F3 lowering produces a “good” /r/ would be helpful in remediation of /r/-production difficulties in children and second language learners. Such a measure would require normalization across speakers. Hagiwara (1995) observed that F3 for /r/ in competent adult speakers was at or below 80% of the average vowel frequencies for a given speaker. In this study, we investigate whether children’s productions start to sound “good” when they lower F3 to the 80% demarcation level or below. Words with /r/ and vowel targets from 20 children with a history of /r/ misarticulation were extracted from acoustic records of speech therapy sessions. Three experienced clinicians judged correctness of /r/ productions. Measured F3’s at the midpoint of /r/ and a range of vowels were compared for these productions. Preliminary findings suggest that the 80% level is a viable demarcation point for good vs. misarticulated articulation of /r/.

Articulatory targets for ultrasound biofeedback determined by tracking regional tongue displacements

Ultrasound biofeedback therapy (UBT) is a significant alternative to traditional therapy for speech disorders, but some users make little progress with the standard feedback display. For segments such as American English /r/, the required tongue movements are so complex that an ultrasound image of the tongue cannot adequately guide speakers to make the right movements. We propose to develop visual display regimes for ultrasound biofeedback that maximize speaker learning. As a preliminary step, we determine displacement ranges of tongue regions (blade, root, and dorsum) known to characterize accurate /r/ production, which will ultimately be used with robust image processing techniques to drive an improved UBT feedback display. Though blade, root, and dorsum movement are known to characterize /r/ production (Boyce et al., 2011; Espy-Wilson et al., 2000; Zhou et al., 2008), excursion of these tongue parts differs across individuals. Here, regional tongue displacements were tracked based on measured motion of...

Perceptual categorization of /r/ for children with residual sound errors

Many studies have found that children with resistant speech sound errors (RSSD) show (1) atypical category boundaries, and (2) difficulty identifying whether their own productions are correct or misarticulated. Historically, perceptual category discrimination tests use synthesized speech representing incremental change along an acoustic continuum, while tests of a child’s self-perception are confined to categorical correct vs. error choices. Thus, it has not been possible to explore the boundaries of RSSD children’s categorical self-perception in any detail or to customize perceptual training for therapeutic purposes. Following an observation of Hagiwara (1995), who noted that typical speakers show F3 values for /r/ between 80% and 60% of their average vowel F3, Hamilton et al (2014) found that this threshold largely replicates adult listener judgments, such that productions above and below the 80% threshold sounded consistently “incorrect” or “correct,” but that productions closest to the 80% threshold w...

Using Landmark Detection to measure effective Clear Speech

While the relationship of clear speaking style to intelligibility under challenging conditions has been established, it is a common observation that some speakers seem to be more intelligible than others for most listeners. In previous work, we have reported that automatic measures based on the technique of Landmark Detection appear to track differences between Clear and Conversational speaking style. One question that remains is whether Landmark measures can be used to predict which speakers are most likely to produce highly intelligible speech. In this study, we took advantage of a set of previously acquired databases to examine correlations between our Landmark-based measures and the Clear Speech productions of highly intelligible speech. Across these databases, we had data on intelligibility for a range of speakers. Results showed that speakers with high overall intelligibility in Clear Speech showed significantly different patterns on Landmark-based automatic measures, compared to speakers with more moderate performance on intelligibility measures. Applications of these results to problems in speech technology, linguistic education and clinical practice will be discussed.

Measuring regional displacements of tongue parts on ultrasound during /ɹ/ articulation

The ability to differentiate movements of the front, back and root portions of the tongue is important to the development of mature speech coordination. This ability is particularly relevant for sounds with complex tongue shapes, such as the American English rhotic approximant /ɹ/ (“r”), but speakers also have a wider scope of coarticulatory opportunities if able to control tongue parts independently [Zharkova, 2012]. In addition, lack of independence in tongue part movement is associated with speech sound disorders [Gibbon, 1999; Gick et al., 2008]. For this study, relative displacements of tongue blade, dorsum, and root were analyzed using MATLAB-based image processing. Regions of interest (ROIs) were drawn for these three areas on ultrasound images during production of /ɑɹ/ by 25 adults. Displacements of each region were measured by tracking of local brightness maxima from images representing /ɑ/ and /ɹ/ production, resulting in ranges of relative blade, dorsum and root displacement associated with nor...

Vowel space in children with residual speech sound disorders

Children with Residual Speech Sound Disorders (RSSD) are considered to have typical speech except for a few misarticulated sounds. During the course of ultrasound biofeedback therapy, parents frequently remark that their children have improved intelligibility, even if the misarticulated sound is not yet successfully remediated. One possible explanation is that therapy improves articulatory precision and expands their articulatory action space. In this study, we collected vowel space measures for 10 RSSD children pre and post 10 sessions of ultrasound therapy. In addition, we collected vowel space measures for 10 typically developing children. Preliminary results suggest a difference between populations. Across RSSD speakers pre-therapy, formant values were variable within phonemic categories; in contrast, formant values were consistent within phonemic categories in typical speakers. Further results of the comparisons between RSSD and typical children as well as for RSSD children’s performance pre and post...

Syllable position effects on perceptual determination of /r/ errors in expert and naive listeners

When ordinary listeners hear words containing a phoneme whose production differs in one feature from the target, the anomaly is often ignored if the word is recognizable. Phonetic research and clinical practice both depend on phonetics training to mitigate these top-down listening effects and enable detection of anomalies. Children with speech sound errors for American English /r/ are believed to have better productions when /r/ occurs at the beginning of a word. However, the shorter duration of /r/ word-initially may be leading clinicians to under-identify errors. While previous studies have shown that clinicians are better than naive listeners at identifying error /r/ in single syllables, they have not assessed identification of /r/ errors in whole words, where position may interact with lexical bias effects. In this study, speech-language pathologists and naive listeners rated children’s natural speech productions of word-initial and word-final/r/ in whole words. These /r/ productions in words were sel...

Articulatory similarity in rhotic sounds: A cross-linguistic comparison

The question of what phonetic quality defines rhotics as a natural class has been debated. Clinical reports indicate that in many languages, rhotics are developmentally late to emerge and subject to errors that are resistant to remediation, suggesting that rhotics may be distinguished by “complexity in articulation.” In American English, the complexity of articulation may derive from its doubly articulated nature; simultaneous palatal and pharyngeal tongue constrictions are consistent features across variants. For other languages with a rhotic liquid, however, it is unclear if this pharyngeal constriction gesture is also an articulatory feature. Using ultrasound, this study compares rhotics from different language families to describe the presence or absence of pharyngeal constriction. Results indicate that pharyngeal constriction is observed across languages in rhotic allophones that have been described as difficult for children to acquire.

How speaker identity interacts with perceptual judgments in children with residual sound errors

Recent research suggests that listeners store complex phonetic representations when learning speech. Encoding fine perceptual details, such as indexical features of the talker’s voice, appears to influence performance in a variety of ways (e.g., processing speed and intelligibility of words in noise). Children with residual sound errors (RSE) for /r/ have been shown to have difficulty judging productions of /r/ from other child speakers along a normalized continuum of third formant values. We hypothesized that children with RSE may make more accurate judgments if they are given stimuli with more familiar indexical characteristics (i.e., their own speech) along the same continuum. In this study, we presented 15 children with a range of stimuli recorded from their own productions as well as productions from other children. In a forced-choice task, children indicated if the word contained a “correct” /r/. Responses to stimuli were compared across children. Initial results suggest that for RSE children, hearing one’s own speech does not improve accuracy in judging the correctness of sounds in words.