The use of ultrasound in the study of articulatory properties of vowels in clear speech

ABSTRACT

Although the acoustic properties of clear speech have been extensively studied, its underlying articulatory details have not been well understood. The purpose of the present study is twofold: To examine the specific articulatory processes of clear speech using ultrasound and to investigate whether and how the type of listener (hard of hearing, normal hearing) and the lexical property of words (frequency) interact in the production of clear speech. To this end, we examined productions of /ɑ/, /æ/ and /u/ from 16 speakers of US English. Overall, our ultrasound results suggested that the tongue’s highest point moved in a direction that exaggerated the three vowels’ phonological features, resulting in an expanded articulatory vowel space for the hard-of-hearing listener and low-frequency words. No interaction was found between the listener and word frequency, suggesting that the effects of word frequency hold constant across the two types of listeners.

KEYWORDS:

• Acoustics
• articulation
• clear speech
• ultrasound
• vowels

Acknowledgements

The author thanks Alana Dust, Elizabeth Flores, Bridget Foley, Kacy Kreger, Leah Theisen and Valerie Stoffer for research assistance.

Declaration of interest

The author reports no conflict of interest.

Funding

This work was supported in part by a seed grant from the College of Health Sciences at the University of Wisconsin-Milwaukee.

Notes

1. Initially, we aimed to examine the effects of two lexical properties, word frequency and phonological neighbourhood density, on the production of clear speech, and therefore, our stimuli were designed to have two levels of neighbourhood density. However, as the study evolves, we decided to leave out phonological neighbourhood density and focus on word frequency effect only. Because existing literature gives mixed findings on the effect of phonological neighbourhood density on clear speech, and because our research was not specifically designed to provide a comprehensive investigation of phonological neighbourhood density, we felt that we were not in a position to adequately address the interaction between this factor and the listener. For the present study, we used only the words from low neighbourhoods, because our preliminary results showed that low-density words were more clearly articulated than high-density words. Because our primary goal is to examine the articulatory characteristics of vowels in clear speech, we chose the condition that facilitated the production of clear speech.Phonological neighbourhood density of each word in the present study was determined from the Hoosier Mental Lexicon, a widely used database of almost 20,000 English words (Nusbaum, Pisoni, & Davis, 1984). Specifically, we used a measure called Neighbourhood Density B (http://neighbourhoodsearch.wustl.edu/Home.asp), which refers to the number of all neighbours that can be created by adding, deleting, or substituting a single phoneme. To our knowledge, there is no available information of mean or distribution of Neighbourhood Density B, so we needed to establish a criterion to determine the high and low neighbourhood density. After a review of the literature, we discovered that neighbourhood density typically ranges from 0 to 40. Therefore, in the present study, any words with Neighbourhood Density B less than 20 were defined to have low neighbourhood density. The average phonological neighbourhood density for low-frequency words used in this study was 11 and that for high-frequency words was 12. In sum, phonological neighbourhood density was matched for low-frequency and high-frequency words in the current study to have low neighbourhood density.

2. One of the anonymous reviewers questioned whether clear speech elicited by the listeners in the video would be comparable to spontaneously produced clear speech. Many studies on clear speech traditionally used the method of instructing participants to read words as if they were talking to a hearing loss, and these studies showed that speakers efficiently switched over to clear speech which resulted in intelligibility benefits. We believe that the video presentation used in the current study may be a more engaging way to elicit clear speech than the method traditionally used. At the same time, some studies have also raised the possibility that the acoustic characteristics of clear speech produced in imagined conditions might differ from those of clear speech produced in more realistic conditions (Hazan & Baker, 2011).

3. We conducted Analysis 1 and 2 using formant frequency values in both Hz and Bark (Traunmüller, 1990). Regardless of which unit of measurement we used, the results were similar and the statistical significance did not change. Here we report the results using formant frequencies in Hz.

Additional information

Funding

This work was supported in part by a seed grant from the College of Health Sciences at the University of Wisconsin-Milwaukee.