Skip to main content
NCBI home page
Journal of Speech, Language, and Hearing Research : JSLHR logoLink to Journal of Speech, Language, and Hearing Research : JSLHR
. 2026 Jul 29;69(9):4056–4068. doi: 10.1044/2026_JSLHR-25-00883

The Effects of Lexical Stress, Lexical Type, and Voicing Context on Symptom Expression in Individuals With Adductor Laryngeal Dystonia

Evangeline Yeh a,✉, Katherine L Marks a,b, Taylor Feaster a, Cara Sauder c, Laura E Toles d, Lauren F Tracy e, J Pieter Noordzij e, Gregory A Grillone e, Ted Mau d, John Paul Giliberto f, Tanya K Meyer f, Albert L Merati f, Tanya L Eadie c,f, Cara E Stepp a,e,g, Saul A Frankford a,h,i
PMCID: PMC13577949  PMID: 42525895

Abstract

Purpose:

The purpose of this study was to examine the effects of lexical stress, voicing context, and lexical type on voice symptoms in individuals with adductor laryngeal dystonia (AdLD).

Method:

Forty-eight individuals with AdLD produced sentences containing a balanced set of words that varied across the linguistic factors of onset voicing, lexical stress, and lexical type. AdLD symptoms (frequency shifts, creaky voice, and voicing breaks) were then identified from the speech acoustics and compared across these factors. An exploratory analysis was performed on recordings from 22 age- and sex-matched controls with the same procedure to assess whether significant effects identified in speakers with AdLD were specific to AdLD.

Results:

Stress and onset voicing had an overall significant effect on symptom expression—that is, stressed syllables elicited more AdLD symptoms than unstressed syllables and syllables beginning with a vowel elicited more symptoms than syllables beginning with an unvoiced consonant. Furthermore, within voiceless consonant-initial syllables, stressed syllables were more likely to elicit AdLD symptoms than unstressed syllables. With respect to lexical type, within unstressed syllables, content words were less likely to contain symptoms than function words. In controls, there were significant effects of onset voicing and an interaction between stress and lexical type but no interaction between stress and voicing context.

Conclusions:

These findings support the well-known effects of voicing onset on symptom elicitation in AdLD and clarify the effects of other linguistic factors such as lexical type and stress. Taken together with previous literature, this work has the potential to advance understanding of the causal mechanisms of task specificity in AdLD.

Supplemental Material:

https://doi.org/10.23641/asha.33050387

Laryngeal dystonia (LD), also known as spasmodic dysphonia, is an action-induced focal dystonia of neurological origin, characterized by spasms of laryngeal muscles that result in voicing irregularities and abnormal speech (Simonyan et al., 2021). The most common form is adductor LD (AdLD; 87% prevalence), which affects the adductor muscles and presents with a strained voice quality paired with intermittent voice breaks during voiced speech sounds (Blitzer et al., 1998; Simonyan et al., 2021). A far less prevalent form is abductor LD (AbLD; 13% prevalence), which affects the abductor muscles and results in a breathier voice quality (Blitzer et al., 1998; Edgar et al., 2001). Individuals with AdLD typically have greater difficulty producing voiced sounds, whereas those with AbLD have challenges with producing voiceless sounds (Cannito et al., 2014). The current gold standard treatment for LD involves injecting botulinum toxin (i.e., Botox) into the intrinsic laryngeal muscles at 2- to 6-month intervals (Blitzer et al., 1998; Ludlow et al., 1988). Notably, speech therapy is not effective for reducing spasms in patients with LD, even when administered in conjunction to regular Botox injections (Murry & Woodson, 1992; Silverman et al., 2012; Whurr & Lorch, 2016). Although the cause and mechanisms underlying LD are unknown, evidence suggests abnormal neural pathways and functions (Simonyan & Ludlow, 2012; Simonyan et al., 2009).

It is notoriously challenging to get an accurate AdLD diagnosis. It often requires visiting three to four physicians to differentiate this disorder from other voice disorders such as primary muscle tension dysphonia (pMTD), resulting in an average diagnostic delay of 5 years (Creighton et al., 2015; Simonyan et al., 2021). One reason for this delay may be low diagnostic reliability, even across experienced clinicians (Ludlow et al., 2018), exacerbated by low prevalence (one in 100,000; Ludlow, 2011). Furthermore, AdLD may co-occur with other voice symptoms such as vocal hyperfunction and/or tremor, which increases diagnostic difficulty (Creighton et al., 2015; Wolraich et al., 2010). There is currently no standardized and objective diagnostic test for AdLD, and diagnosis is carried out through the collaboration of otolaryngologists, speech-language pathologists, and neurologists (Whurr & Lorch, 2016).

A unique aspect of LD that may aid diagnosis is that its symptoms appear to be task specific. This means that the voice symptoms associated with LD occur most frequently during volitional speech, whereas voicing for laughing, crying, whispering, humming, and singing are largely typical (Bloch et al., 1985; Sapienza et al., 1999). It is also context specific, such that symptoms occur more frequently in sentences containing certain sounds (Ludlow et al., 2008; Sapienza et al., 1999). Ludlow et al. (2008) created a comprehensive LD diagnostic procedure based on a survey of expert scientists and clinicians that included three steps: screening questions (possible diagnosis of LD), speech examination (probable diagnosis of LD), and a nasolaryngoscopy (definite diagnosis of LD). For the speech examination tier of the assessment, they compiled sentences that were intended to elicit specific adductor or abductor voice symptoms and asked participants to repeat the sentences in both a normal speaking voice and a whisper. A speech-language pathologist rated their voice using a list of symptoms, which included vowel breaks, to determine if AdLD, AbLD, or hyperfunctional voice (i.e., pMTD) was present (Ludlow et al., 2008). Results in this study suggested that speech examination may help differentiate AdLD from pMTD and may be a useful diagnostic tool to detect AdLD in individuals (Ludlow et al., 2008). However, there are limitations to this three-tiered approach to LD diagnosis, such as the poor interrater reliability among specialized medical professionals in classifying LD based on speech examinations and nasolaryngoscopy video recordings (Ludlow et al., 2018).

Developing more sensitive stimuli may require determining stimulus characteristics that increase or decrease the prevalence of symptoms. Several linguistic factors have been shown to influence the presence of symptoms in speakers with AdLD, including voicing, phoneme context, manner of articulation, prosodic stress, and lexical type (Cannito et al., 2014; Erickson, 2003; Frankford et al., 2025; Froeschke, 2020; Leonard & Kendall, 1999; Ludlow & Connor, 1987; Reid & Nobriga, 2023; Roy et al., 2007, 2013; Whurr & Lorch, 2016). As mentioned previously, the main symptom distinguishing the adductor and abductor types of LD is that individuals with AdLD have more difficulty with voiced-loaded sentences, whereas individuals with AbLD have more difficulty with voiceless loaded sentences. Cannito et al. (2014) evaluated the sentence duration and voice quality of AdLD and AbLD participants across these types of sentences and found that individuals with AdLD were judged by voice clinicians to have poorer voice quality when they read sentences with voiced-loaded words compared to those with voiceless loaded words. These and other similar findings (Leonard & Kendall, 1999; Roy et al., 2007, 2013, 2024) suggest that examining the voice quality of speakers in voiced-loaded sentences may be useful in differentiating AdLD from other voice disorders. Furthermore, certain contexts also seem to elicit more symptomatic speech in participants with AdLD, such as vowel–vowel word transitions (Ludlow & Connor, 1987). In a study of speakers with AdLD, participants produced fewer repeated /a/ sounds in 5 s than /pa/ sounds, and when compared with controls, the participants with AdLD had similar phonation times but longer times between phonation during /a/ repetition and not /pa/ repetition (Ludlow & Connor, 1987). These findings suggest that speakers with AdLD have difficulties with vowel–vowel contexts, mainly in the form of protracted glottal closure (Ludlow & Connor, 1987). Interestingly, typical speakers exhibit increased glottalization, including glottal stops and glottal fry, in word-initial vowels preceded by vowels (Dilley et al., 1996), suggesting that these contexts require increased glottal activity or control that may trigger spasms in AdLD.

Recent papers have also explored the effects of manner and place of articulation in AdLD symptom expression. Frankford et al. (2025) found that when speakers with AdLD read an all-voiced sentence aloud, vowels and approximants elicited the most symptoms, whereas nasal consonants and obstruents (stops and fricatives) elicited fewer symptoms. This indicates that even in an all-voiced sentence, symptoms do not occur evenly or equally across sounds but depend on phonetic context (Frankford et al., 2025). Similar findings were reported in a separate study, in which results demonstrated that words beginning with liquid consonants (such as /l/ or /r/) were more consistently perceived as containing a spasm than words beginning with fricatives (Reid & Nobriga, 2023). Additionally, Reid and Nobriga (2023) found that words beginning with vowels, bilabials, alveolars, and the glottal /h/ had more symptoms than those beginning with linguodental sounds. Thus, the place and manner of articulation may influence the degree of symptoms in the voice of individuals with AdLD.

Other factors affecting the likelihood of symptoms in the speech of individuals with AdLD include the syntactic complexity of a sentence and the proportion of content versus function words (lexical type). One study found that syntactic complexity (more vs. less complex sentences) increased the probability of symptom appearance (Erickson, 2003). A separate study found that sentences with greater lexical density (a greater proportion of content words yielding greater cognitive or information load) or containing lower lexical frequency words (e.g., industry-specific technical terms) both increased the probability of symptom expression in speakers with AdLD (Froeschke, 2020). Furthermore, Froeschke (2020) found a statistically significant difference in voice breaks based on lexical type—content items, such as nouns or verbs, yielded more symptoms than function items, such as prepositions, pronouns, adverbs, and articles. Content words convey the substance of a sentence and are essential to ensure that the message is conveyed, whereas function words support the message but are less essential (Segalowitz & Lane, 2000). Since LD is a task-specific voice disorder, with symptoms mainly occurring in speech contexts, it has been hypothesized that increased symptom appearance is related to speech with increased value. Specifically, more important or meaningful words (i.e., content words) may elicit more symptoms than less meaningful words (i.e., function words; Ludlow & Connor, 1987). Alternatively, the effects of lexical type on symptoms in AdLD may be confounded by the effects of linguistic stress (Reid & Nobriga, 2023). Reid and Nobriga (2023) found that both stressed syllables and content words had significantly more consistently perceived AdLD spasms compared with unstressed syllables and function words, respectively. However, there was a large overlap between stressed words and content words because sentential stress typically occurs on more meaningful parts of the sentence (Reid & Nobriga, 2023). It is possible, then, that the increased laryngeal motor control requirements for modulating stress and other prosodic cues of speech may belie the influence of these linguistic factors on symptoms in AdLD.

A drawback to some of these previous studies was that they used speech stimuli of convenience as they were already standard sentences used in clinical diagnosis (Frankford et al., 2025; Froeschke, 2020; Reid & Nobriga, 2023), rather than developing stimuli that are balanced across levels of the factors of interest. Additionally, others coded or analyzed data at the level of whole sentences (Cannito et al., 2014; Erickson, 2003), precluding more fine-grained analysis of speech segments. Furthermore, each of these studies examined these factors in isolation, without accounting for how they may confound each other (e.g., content words are more likely to be stressed than function words). Therefore, the present study sought to investigate differences in symptom expression in participants with AdLD across three different factors—syllable onset voicing (vowel onset vs. voiceless consonant onset), lexical stress (stressed vs. nonstressed), and lexical type (content vs. function)—using balanced stimulus sentences to control for the effects of the other variables. Based on the prior literature, we hypothesized that stressed syllables would elicit greater symptoms than nonstressed syllables, vowel onsets would elicit greater symptoms than voiceless consonant onsets, and content words would elicit greater symptoms than function words in speakers with AdLD. We also examined whether the effects of these three factors varied across levels of the other factors by including interaction terms in the models. Finally, a similar analysis was completed in a small set of age- and sex-matched control participants to assess whether effects found in speakers with AdLD were specifically related to AdLD symptom expression rather than typical glottalization patterns.

Method

All participants completed written consent in compliance with the Boston University Institutional Review Board (Protocol No. 2625), University of Washington Institutional Review Board (Protocol No. 36181), or The University of Texas Southwestern Institutional Review Board (Protocol No. STU-2022-0388).

Participants

Forty-eight individuals (35 women, 13 men) diagnosed with AdLD between 32 and 80 years of age (women average = 62.8 years, SD = 12.8; men average = 54.6 years, SD = 13.9) were participants in this experiment and were tested at three different sites (nine tested at Boston University [BU], 10 tested at The University of Texas Southwestern Medical Center [UTSWMC], and 29 tested at the University of Washington [UW]). Speakers exhibited a range of AdLD severity and were all symptomatic at the time of recording, either because they were recorded less than 2 weeks prior to receiving Botox injections (with a dosage frequency range of once a month to twice a year) or they were not actively receiving Botox (n = 1). All participants were diagnosed by a board-certified otolaryngologist and reported having no other history of speech, language, and hearing disorders or neurological conditions. All participants were speakers of English. Speakers judged by the first and last authors as having a non–American English accent were excluded from this study. An experienced voice clinician assessed the speech samples of all participants with AdLD and rated their voices based on the Consensus Auditory-Perceptual Evaluation of Voice (Kempster et al., 2009), which assesses features such as overall severity of dysphonia, roughness, breathiness, and strain on a scale of 0–100. In addition, the clinician rated the “severity of spasm” on a scale of 0–100. The clinician's ratings yielded a range of 9.95—76.6 for overall severity of dysphonia and 0–75.7 for spasm. In addition, the clinician assessed whether vocal tremor was present in each speaker and whether they were certain of the presence of vocal tremor in their assessment. For the three participants for whom the clinician was uncertain, two additional voice-specialized clinicians rated the same samples. For both overall severity of dysphonia and spasm, intrarater reliability was excellent (overall severity: intraclass correlation coefficient (A,1) [ICC(A,1)] = 0.90, spasm: ICC(A,1) = 0.93). Speakers identified as having vocal tremor by at least two clinicians were considered to have co-occurring tremor. Eighteen participants were judged as having co-occurring tremor, and 30 were labeled as not having co-occurring tremor.

Twenty-two additional control participants (16 women, six men) between the ages of 34 and 75 years (Mwomen = 60.1 years, SD = 11.9; Mmen = 60.3 years, SD = 9.3) with no history of voice, speech, or neurological disorders were also tested (all at BU) to determine whether the results of our primary analysis were specific to speakers with AdLD. Similar to the speakers with AdLD, all control participants spoke English without a detectable non–American English accent.

Recording Procedure

Each participant read a set of eight sentences as part of a larger speech protocol. Recordings were collected in either a quiet room (n = 34) or a sound-attenuating booth (n = 36) using a condenser microphone (UW: AKG C420; UTSWMC: AKG C520; BU: Shure MX153) placed near the mouth (UW and BU: 7 cm, 45° from the mouth; UTSWMC: 5 cm, 45° from the mouth). Signals were digitized at a sampling frequency of 44,100 Hz and recorded to either a digital audio workstation or a Zoom H6 audio recorder (Zoom). The stimulus sentences used in this experiment (see Table 1) were displayed to participants in print. Following verbal instructions from the researcher, participants read the stimulus sentences aloud at a natural pitch and loudness at their own pace. When necessary, the researcher would prompt participants to repeat sentences if they mispronounced or misread words. In cases in which sentences were repeated due to errors, only the repeated sentences were included in our analysis.

Table 1.

Stimulus sentences.

Vowel-onset target words Voiceless consonant–onset target words
1. He enteredC the meadow acrossF the way almostF as soon as he achievedC his goal. 5. The superbC turkey somehowF flew himselfF to the farmerC.
2. We onlyF see AndyC alongF the ascentC. 6. We hardlyF saw HannahC go towardF the salonC.
3. The silly emuC evenF walked slowly aroundF the irateC group. 7. Would you perhapsF be sorryC to confrontC a secondF shark?
4. Row overF the enclosedC bay insteadF of the openC water. 8. She herselfF grew somewhatF frustrated but remained so civilC and calmly composedC.
Open in a new tab

Note. Underlined = target words; bold = stress. C denotes a content word. F denotes a function word.

Stimulus Sentences

Eight sentences were created as stimuli for this experiment (see Table 1), and four target words were embedded in each sentence. Because prior work has shown that voiced-loaded sentences and voiced word onsets elicit more symptoms in speakers of AdLD than voiceless loaded sentences and voiceless word onsets (Cannito et al., 2014; Frankford et al., 2025; Ludlow & Connor, 1987), the target words in this experiment included vowel-onset and voiceless consonant-onset syllables. In four of the sentences, all target words started with a vowel and were preceded by a word that ended in a vowel, creating vowel–vowel word transitions (e.g., “see Andy”). This context was chosen to maximize the opportunity for symptoms given that vowel–vowel junctions are conducive to glottalization (Dilley et al., 1996) and are associated with greater symptoms in speakers with AdLD (Ludlow & Connor, 1987). In the remaining sentences, each target word began with an unvoiced consonant and was similarly immediately preceded by a word ending in a vowel (e.g., “be sorry”). To complement Reid and Nobriga's (2023) findings on prosodic stress as a factor influencing voice symptoms, this study examined lexical stress within bisyllabic words. This allowed us to balance stressed and unstressed first syllables between content and function words. Each sentence contained two content words and two function words, one with lexical stress on the first syllable and one with lexical stress on the second syllable. Only the first syllable of each target word was analyzed (for more information, see the Data Extraction section below). An example of a sentence with vowel-onset target words is “He entered the meadow across the way almost as soon as he achieved his goal.” In this sentence, “entered” and “achieved” are content words, whereas “across” and “almost” are function words. At the same time, “entered” and “almost” have a stressed–unstressed (trochaic) pattern, whereas “achieved” and “across” have an unstressed–stressed (iambic) pattern. The complete set of stimuli used in this experiment is included in Table 1.

Symptom Labeling

Sapienza et al. (1999) outline three main symptoms of AdLD: phonatory breaks, aperiodic segments (creaky or noisy voice), and frequency shifts during vowel production. A trained technician identified and manually labeled these symptoms in each set of stimuli from each speaker using Praat (Version 6.1.03; Boersma & Weenink, 2019). AdLD symptom identification followed guidelines from Sapienza et al. (1999), Sapienza et al. (2002), and Marks et al. (2022). Phonatory breaks were defined as pauses in voicing longer than 50 ms (Sapienza et al., 2002) or by a pause in voicing in segments where voicing would be expected (within a syllable that requires voicing, Sapienza et al., 1999, visualized in Figure 1A). Additionally, to capture other relevant symptoms, pauses in voicing between words that were perceptually noticeable as breaks were labeled as such. Frequency shifts were defined as changes of 50 Hz in fundamental frequency (F0) or more within 50 ms (Sapienza et al., 1999, p. 131, illustrated in Figure 1B). Pitch settings were customized for each speaker prior to symptom labeling in order to accurately track their F0 and any pitch shifts that might occur. The pitch tracking algorithm used in Praat was the autocorrelation method. Pitch range was adjusted to each participant's voice as follows: Labelers found a segment of the recording where the pitch tracking was representative of their modal voice and obtained a mean F0 for that portion. Labelers set the minimum pitch to be more than half of the obtained mean F0 and the maximum pitch to be less than double the obtained mean F0 in order to remove mistracked pitch segments. In the case where a participant's voice was too dysphonic to have any consistent segments of modal voice, labelers used the general guideline of 150–300 Hz for female voices and 75–200 Hz for male voices. Instances of creaky voice were labeled for several types of atypical voicing, including aperiodic speech (Sapienza et al., 1999, visualized in Figure 1C), irregular F0 (i.e., segments containing non repetitive cycles), vocal fry, multiply pulsed voice (F0 irregularity which involves alternating longer and shorter pulses), and any general low F0 (significantly lower F0 than their modal voice; Keating et al., 2015; Marks et al., 2022; Sapienza et al., 1999). Following methods from Marks et al. (2022), all symptoms were labeled in Praat using both waveform and spectrographic visualization and auditory perception. If multiple symptoms appeared within a speech segment, the technician labeled all possible symptoms. The trained technician relabeled 19% (13/70) of participant recordings selected at random to determine intrarater reliability. A second trained technician labeled 19% (13/70) of participant recordings for symptoms to determine interrater reliability.

Figure 1.

The image displays three panels, A, B, and C, stacked vertically, illustrating Adductor laryngeal dystonia (AdLD) symptoms identified in Praat from one individual with AdLD, each showing a different voice symptom with its corresponding waveform and measurements. Panel A presents a waveform demonstrating a phonatory break. The waveform begins with clear, periodic oscillations, then transitions into a flat line, indicating a cessation of phonation, before resuming periodic oscillations. A light red shaded area highlights this flat line segment, which is labeled Phonatory Break and measured to be 125 milliseconds in duration. Panel B depicts a frequency shift. The upper part of the panel displays a waveform that initially shows periodic oscillations. Within a light red shaded area, the oscillations become more compressed and then gradually spread out, indicating a change in frequency. This segment is labeled Frequency Shift and has a duration of 32 milliseconds. Below the waveform, a blue line represents the fundamental frequency trace. This trace shows a decrease from 260 Hertz at the beginning of the shaded region to 197 Hertz at its end, signifying a 63 Hertz drop in frequency over the 32 millisecond period. Panel C shows a waveform illustrating creaky voice. The waveform starts with regular, periodic oscillations, then changes to an irregular, less periodic, and somewhat noisy pattern within a light red shaded area. This segment is labeled Creak. Following this, the waveform returns to regular, periodic oscillations.

Open in a new tab

Adductor laryngeal dystonia (AdLD) symptoms identified in Praat from one individual with AdLD. Panel A: waveform and time boundaries for an example of a phonatory break lasting 125 ms. Panel B: waveform, fundamental frequency trace, and time boundaries of a frequency shift showing a change of 63 Hz in 32 ms. Panel C: waveform and time boundaries for an example of a segment of vocal “creak.” FS = frequency shift.

Speech Segment Boundary Marking

Word and phoneme boundaries were first identified using the forced alignment toolkit Penn Phonetics Lab Forced Aligner (P2FA; Yuan & Liberman, 2008). Using phoneme boundaries from the P2FA alignment toolkit as guidance, syllable boundaries for each of the target syllables (underlined in Table 1) were added by the technician to mark segments in which data extraction occurred. The boundaries were then adjusted manually by a trained technician. The trained technician adjusted 19% of the boundaries in each of the sentences twice to determine intrarater reliability. A second technician labeled syllable boundaries for a random 19% (13/70) of participant recordings to determine interrater reliability. Segment boundary labeling was performed following the completion of all symptom labeling and without reference to this prior labeling to minimize potential bias in the labeling tasks.

Data Extraction

After labeling both AdLD symptoms and speech segment boundaries, an automated MATLAB script determined whether any symptom overlapped with the 32 target syllables for all 48 AdLD participants and 22 control participants. Thus, the recordings were coded for each syllable using a binary system: 1 for occurrence for any symptom in the syllable, based on the segment boundaries marked, and 0 for lack of occurrence of any symptom. As in Marks et al. (2024), symptoms needed to overlap with the syllable by at least 15 ms or occur entirely within the time boundaries of the syllable (even if less than 15 ms) for the symptom to be associated with the syllable. The data set contained 32 target syllables from each of the 48 speakers with AdLD and 22 control speakers, resulting in a total of 2,240 potential data points. Any syllables that were “errors” (e.g., misread, not read, loss of voicing context) were removed from analysis: 141 (6.3%) were not included in the final analysis due to incorrect voicing (one), general misread words (inclusive of incorrect stress; 43), word omission (two), dropped consonants (eight), or loss of voicing context (87). Loss of context meant that either the word preceding the target syllable was misread or there was a pause > 200 ms, such that the vowel–vowel or vowel–unvoiced consonant context was not achieved.

Statistical Analyses

Intra- and interrater reliability for phoneme boundary labeling was evaluated using two methods. The first method examined the ICC (ICC(A,1) for absolute agreement; McGraw and Wong, 1996) of the durations of each target syllable. The second method evaluated the mean and median absolute error in boundary timing for all target syllables—that is, central tendencies of how different syllable onset and offset times were across repeated ratings. Intra- and interrater reliability for symptom labeling was evaluated using Cohen's kappa as a measure of the agreement within/between technicians on whether a symptom was present for each target syllable, using syllable boundaries from the first technician.

Primary Analysis in Speakers With AdLD

To evaluate effects of lexical stress, onset type, and lexical type as well as their second- and third-order interactions, a logistic linear mixed-effects model with random intercepts for each participant (similar to Frankford et al., 2025) was fit. All analyses were conducted using the lme4 package in R (Version 4.1.2; The R Foundation)/RStudio (Version 2025.05.0; Posit, PBC). In the main linear mixed-effects model, the binary outcome variable was the presence or absence of a symptom for the first syllable of each target word. Lexical stress, voicing, lexical type, and their interactions were modeled as fixed effects. Syllable duration was modeled as a fixed-effects covariate to account for differences in the opportunity for a symptom to occur across target words and speakers. Random effects consisted of random intercepts for each speaker. Significant fixed effects were followed up with post hoc tests statistically comparing estimating marginal means using the Tukey method to correct for multiple comparisons (package emmeans, Version 1.7.3).

Secondary Analysis in Control Speakers

One of the reasons for using vowel–vowel word transitions in this study was evidence of increased glottalization (e.g., creaky voice and glottal stops) for typical speakers in this context (Dilley et al., 1996). Because of this, changes in the probability of AdLD symptoms of creak or voice breaks across voicing categories could reflect either AdLD pathology or an increase in typical glottalization. Similarly, changes in symptoms across lexical stress and lexical type may reflect patterns of glottalization not specific to AdLD. To disentangle these effects, a secondary analysis was carried out in the group of control speakers. This analysis examined whether significant effects of voicing, lexical stress, lexical type, and their interactions were present only in speakers with AdLD or whether these effects could be found in typical speakers as well. This was also carried out using a logistic linear mixed-effects model with voicing, lexical type, lexical stress, and their interactions as fixed effects; a fixed-effects covariate of syllable duration; and random intercepts for each speaker.

Results

Intra- and interrater reliability analysis for phoneme boundary labels and symptom labels was carried out for target syllables from 19% (13/70) of participants, excluding syllables that contained errors (described above). For syllable boundaries, the between-raters ICC absolute agreement (Type A-1) for syllable duration was 0.93, and the within-rater ICC absolute agreement (Type A-1) for syllable duration was 0.96. Because this measure only accounts for agreement of syllable duration and not agreement of syllable onsets and offsets, the mean absolute difference of syllable onsets and offsets (separately) within and between raters was also examined. Collapsing across onsets and offsets, within-rater mean absolute difference was 10.1 ms, and median absolute difference was 1.8 ms; between-raters mean absolute difference was 14.4 ms, and median absolute difference was 8.8 ms. Using the primary labeler's syllable boundaries, there was moderate within-rater agreement on which syllables contained symptoms (Cohen's κ = .60; 82% overall agreement) and substantial between-raters agreement (Cohen's κ = .76; 88% overall agreement). We note that the result of interrater reliability being greater than that of intrarater reliability is unusual but accurately reflects the completed symptom labeling.

Primary Analysis in Speakers With AdLD

A plot of the raw proportion of symptoms found in each combination of factors (mean and standard error across participants) can be found in Figure 2. In order to assess the effect of each linguistic factor in symptom expression, generalized linear mixed-effects models were fit to the data. Although prior studies investigating the effects of language on symptoms using this method included speech sound duration as a covariate to account for differences in the opportunity for symptoms to occur (Frankford et al., 2025; Marks et al., 2024), a concern was that syllable duration would confound stress since unstressed syllables tend to be shorter than stressed syllables. Indeed, there was a significant difference in duration between stressed and unstressed syllables (stressed syllables: M = 251 ms, SD = 84; unstressed syllables: M = 178 ms, SD = 89; t(1432) = 18.12, p < .001). Therefore, two versions of the model were run—one with duration as a covariate and another without—to determine whether there was an expected effect on the results (a reduction in the effect of stress). When comparing both models (with and without duration), stress remained significant with an effect in the same direction. Additionally, the model with duration fit the data better based on Akaike Information Criterion. Therefore, only the results from the model including the duration covariate are reported here (see Supplemental Material S1 and Supplemental Material S2 for the alternative model).

Figure 2.

A bar chart displays the raw proportion of tokens, expressed as a percentage, containing symptoms for two groups: AdLD (adductor laryngeal dystonia), represented by blue bars, and Control, represented by orange bars. The y-axis is labeled Raw Proportion (in percent) and ranges from 0 to 100 percent. The x-axis is divided into two main sections: Unvoiced Onsets and Vowel Onsets. Each of these sections is further divided into Unstressed and Stressed conditions, and within each condition, there are bars for F (function) and C (content) words. Error bars indicate the standard error of the mean across speakers. For Unvoiced Onsets, the AdLD group consistently shows higher proportions than the Control group. In the Unstressed condition, AdLD shows approximately 25 percent for Function words and 23 percent for Content words, while the Control group shows very low proportions, around 0 to 1 percent for both. In the Stressed condition for Unvoiced Onsets, AdLD shows about 42 percent for Function words and 48 percent for Content words. The Control group remains low, with approximately 2 percent for Function words and 6 percent for Content words. For Vowel Onsets, both groups generally show higher proportions than for Unvoiced Onsets, with AdLD still significantly higher than Control. In the Unstressed condition, AdLD shows approximately 65 percent for Function words and 52 percent for Content words. The Control group shows about 30 percent for Function words and 15 percent for Content words. In the Stressed condition for Vowel Onsets, AdLD shows around 67 percent for Function words and 65 percent for Content words. The Control group shows approximately 27 percent for Function words and 30 percent for Content words.

Open in a new tab

Raw proportion of tokens (expressed as a percent) containing symptoms for each combination of factors in speakers with adductor laryngeal dystonia (AdLD) and controls. Errors bars indicate the standard error of the mean across speakers. F = function; C = content.

The results of this model can be found in Table 2 and Figure 3. Significant main effects included voicing (OR = 4.49, 95% confidence interval [CI] [2.68, 7.52], p < .001) and stress (OR = 0.19, 95% CI [0.19, 0.53], p < .001), but not lexical type (OR = 0.65, 95% CI [0.41, 1.04], p = .075). However, these main effects were modulated by interaction terms. Significant interactions were found between voicing and stress (OR = 2.43, 95% CI [1.19, 4.95], p = .014) and between lexical type and stress (OR = 2.38, 95% CI [1.14, 4.98], p = .021; see Figure 4). The interaction between lexical type and voicing and the three-way interaction among voicing, lexical type, and stress term were not statistically significant. The covariate syllable duration was found to be statistically significant (OR = 1.83, 95% CI [1.50, 2.23], p < .001).

Table 2.

Results of the primary analysis in speakers with adductor laryngeal dystonia.

Fixed effects
Variables OR 95% CI z value p
Voicing 4.49 [2.68, 7.52] 5.71 < .001*
Lexical type 0.65 [0.41, 1.04] −1.78 .074
Stress 0.32 [0.19, 0.53] −4.30 < .001*
Duration 1.83 [1.50, 2.23] 5.94 < .001*
Voicing × Lexical Type 1.77 [0.89, 3.53] 1.62 .104
Voicing × Stress 2.43 [1.19, 4.95] 2.45 .014*
Lexical Type × Stress 2.38 [1.14, 4.98] 2.30 .021*
Voicing × Lexical Type × Stress 0.75 [0.27, 2.08] −0.55 .586
Open in a new tab

Note. CI = confidence interval.

*

p < .05.

Figure 3.

This image presents a series of point plots with error bars, illustrating the estimated probability of symptom expression in speakers with adductor laryngeal dystonia. The y-axis represents the Estimated Symptom Probability ranging from 0 to 75 percent. The x-axis categorizes Lexical Type into Content and Function words. The overall plot is divided into two main panels based on voicing context: Unvoiced on the left and Vowel on the right. Within each panel and lexical type, two conditions are compared: Stressed syllables, marked by red circles, and Unstressed syllables, marked by blue circles, as detailed in the legend. Vertical error bars extending from each point indicate the 95 percent confidence intervals for the estimated probabilities. In the Unvoiced panel, for Content words, the estimated symptom probability for stressed syllables is approximately 35 percent, which is notably higher than for unstressed syllables, estimated at about 15 percent. For Function words within the Unvoiced context, stressed syllables show an estimated probability of around 28 percent, while unstressed syllables are slightly lower at approximately 20 percent. Across both Content and Function lexical types in the Unvoiced panel, stressed syllables consistently exhibit a higher estimated symptom probability compared to unstressed syllables. Shifting to the Vowel panel, the estimated symptom probabilities are generally much higher than those observed in the Unvoiced panel. For Content words in the Vowel context, stressed syllables have an estimated probability of about 70 percent, which is marginally higher than unstressed syllables at approximately 65 percent. For Function words in the Vowel context, unstressed syllables show the highest estimated probability at around 78 percent, which is slightly greater than stressed syllables at approximately 72 percent.

Open in a new tab

Estimated probability of symptom expression in speakers with adductor laryngeal dystonia across stress, voicing, and lexical type, accounting for duration, reverse transformed from the logit scale to the probability scale. Error bars indicate 95% confidence intervals.

Figure 4.

A point plot displays the estimated probability of symptom expression in speakers with adductor laryngeal dystonia across voicing and stress. The y-axis represents the Estimated Symptom Probability, ranging from 0 to 80 percent. The x-axis represents Onset Voicing, with two categories: Unvoiced and Vowel. Data points are differentiated by stress condition: Stressed is shown in red, and Unstressed is shown in blue. In the Unvoiced onset voicing condition, the estimated symptom probability for Stressed words is approximately 30 percent, with 95 percent confidence intervals extending from about 22 percent to 40 percent. For Unstressed words in the Unvoiced condition, the estimated probability is approximately 17 percent, with 95 percent confidence intervals extending from about 10 percent to 25 percent. A double asterisk above these points indicates that the difference between Stressed and Unstressed words in the Unvoiced condition is statistically significant (p less than 0.01). In the Vowel onset voicing condition, the estimated symptom probability for Stressed words is approximately 70 percent, with 95 percent confidence intervals extending from about 62 percent to 78 percent. For Unstressed words in the Vowel condition, the estimated probability is approximately 72 percent, with 95 percent confidence intervals extending from about 65 percent to 80 percent. The annotation n.s. (not significant) above these points indicates that the difference between Stressed and Unstressed words in the Vowel condition is not statistically significant. A large black double-sided arrow, marked with triple asterisks, extends diagonally from the lower left (Unvoiced condition) to the upper right (Vowel condition). This arrow signifies that all comparisons between the Unvoiced and the Vowel condition are highly statistically significant (p less than 0.001).

Open in a new tab

Estimated probability of symptom expression in speakers with adductor laryngeal dystonia across voicing and stress, collapsed across voicing conditions and accounting for duration. Estimates are reverse transformed from the logit scale to the probability scale. Error bars indicate 95% confidence intervals. The double-sided arrow indicates that all comparisons between the unvoiced and the vowel condition are significant. Error bars indicate 95% confidence interval. **p < .01. ***p < .001. n.s. = not significant.

Significant interactions were explored with post hoc pairwise comparisons. For the Voicing × Stress interaction in speakers with AdLD (see Figure 4), marginal means were estimated after collapsing across lexical types, and pairwise comparisons were made using a Tukey correction to compare the four combinations of voicing and stress. All differences between unvoiced and vowel onsets were statistically significant (all p < .001). Among target syllables with unvoiced onsets, stressed syllables had a greater chance of eliciting symptoms than unstressed syllables (stressed: 0.299, unstressed: 0.172, p = .002, OR = 0.49, 95% CI [0.29, 0.81]). However, in target syllables beginning with a vowel, there was no effect of stress (stressed: 0.718, unstressed: 0.724, p = .999, OR = 1.03, 95% CI [0.62, 1.70]). Post hoc analysis of the Stress × Lexical Type interaction term (see Figure 5) showed that, in unstressed syllables, function words yielded more symptoms than content words (function: 0.498, content: 0.356, p = .011, OR = 1.80, 95% CI [1.10, 2.93]). No effect of lexical type was found for stressed syllables (function: 0.493, content: 0.528, p = .852, OR = 0.87, 95% CI [0.55, 1.36]). Breaking this interaction down by lexical type, although there was no significant effect of stress among function words (stressed: 0.493, unstressed: 0.498, p = 1, OR = 1.02, 95% CI [0.60, 1.72]), stressed content words showed greater symptoms than unstressed content words (stressed: 0.528, unstressed: 0.356, p = .001, OR = 0.49, 95% CI [0.30, 0.81]).

Figure 5.

The image is a point range plot illustrating the estimated symptom probability for two lexical types, Content and Function words, under Stressed and Unstressed conditions. The y-axis, labeled Estimated Symptom Probability, ranges from 30 to 60 percent. The x-axis, labeled Stress, has two categories: Stressed words and Unstressed words. Red points and error bars represent Content words, while blue points and error bars represent Function words. Error bars indicate 95 percent confidence intervals. In the Stressed condition, the estimated symptom probability for Content words is approximately 52.5 percent, with its confidence interval spanning from about 42 to 63 percent. For Function words in the Stressed condition, the estimated probability is around 49.5 percent, with its confidence interval from approximately 39 percent to 59 percent. There is no significant difference (n.s.) between Content and Function words in the Stressed condition. In the Unstressed condition, the estimated symptom probability for Content words is approximately 36 percent, with its confidence interval from about 27 percent to 45 percent. For Function words in the Unstressed condition, the estimated probability is around 49.5 percent, with its confidence interval from approximately 39 to 60 percent. There is a significant difference (p less than 0.05) between Content and Function words in the Unstressed condition, with Function words showing a higher probability. Comparing across stress conditions, Content words shows a highly significant decrease (p less than 0.01) in symptom probability from Stressed to Unstressed. Conversely, Function words shows no significant difference (n.s.) in symptom probability between the Stressed and Unstressed conditions.

Open in a new tab

Estimated probability of symptom expression in speakers with adductor laryngeal dystonia across lexical type and stress, collapsed across voicing conditions and accounting for duration. Estimates are reverse transformed from the logit scale to the probability scale. Error bars indicate 95% confidence intervals. Error bars indicate 95% confidence interval. *p < .05. **p < .01. n.s. = not significant.

Secondary Analysis in Control Speakers

To determine whether changes in symptoms were related to AdLD (and not consistent with glottalization found in typical speakers), a follow-up analysis in age- and sex-matched control speakers was completed. Because the control group produced no evidence of glottalization (i.e., “symptoms”) for voiceless, unstressed, function words (and thus no variability across participants), there was difficulty uniquely estimating all terms in the full model. To resolve this, the three-way Voicing × Lexical Type × Stress interaction term was removed from the model. The complete results can be found in Supplemental Materials S3, S4, and S5. Similar to speakers with AdLD, this reduced model showed significant effects of voicing (β = 2.37, SE = 0.53, p < .001) and a significant Lexical Type × Stress interaction (β = 1.28, SE = 0.56, p = .021). Other model terms that were significant in the AdLD model and not significant in this model included the main effect of stress (β = −1.85, SE = 1.12, p = .10) and the Voicing × Stress interaction (β = 1.04, SE = 1.15, p = .37).

Post hoc analysis of the Stress × Lexical Type interaction term for control speakers showed that, in stressed syllables, content words yielded more symptoms than function words on average (function: 0.093, content: 0.030), but this was not statistically significant (p = .20, OR = 0.305, 95% CI [0.064, 1.45]). There was very little difference between content words and function words in the unstressed condition (p = 1, OR = 1.103, 95% CI [0.197, 6.17]).

Discussion

In this study, we used linear mixed-effects models to examine whether AdLD symptom prevalence varied depending on (a) syllable onset voicing, (b) lexical stress, and (c) lexical type, controlling for the effects of the other variables. We hypothesized that stressed syllables would elicit greater symptom expression than nonstressed syllables, vowel onsets would elicit greater symptom expression than voiceless consonants onsets, and content words would elicit greater symptom expression than function words in speakers with AdLD. Our findings indicated that syllable onset voicing and lexical stress, but not lexical type, were linguistic factors that significantly affected symptom expression in AdLD speakers. However, these effects were modulated by significant interactions between onset voicing and stress and between lexical type and stress. Post hoc testing revealed that, overall, vowel-initial syllables were more likely to result in symptoms than voiceless consonant-initial syllables. Additionally, stressed syllables only led to more symptoms than unstressed syllables in voiceless consonant-initial contexts. Furthermore, stress was found to have a significant effect on symptoms in content words but not function words. Intriguingly, unstressed syllables in content words (e.g., irate) were less likely to contain symptoms than unstressed syllables in function words (e.g., around), but no main effect of lexical type was found for stressed syllables. We conducted a follow-up analysis investigating the effects of these factors on acoustic features of AdLD symptoms in typical speakers to determine whether these effects could have been due to typical differences in glottalization across these contexts. Results of this analysis revealed a significant effect of voicing onset and a significant interaction between lexical type and lexical stress. These effects will be discussed in more detail in the sections that follow.

Lexical Stress

In the present experiment, we found a main effect of lexical stress on AdLD symptoms in speech, such that stressed syllables overall had more symptoms than unstressed syllables. This is consistent with previous literature—Reid and Nobriga (2023) found that stress had a moderate effect on perceived symptoms where stressed words were more consistently rated as containing spasms than unstressed words. It is worth noting that, whereas Reid and Nobriga analyzed prosodic stress on one-syllable words, the present study analyzed lexical stress in two-syllable words. However, in this study, we found that the effects of stress were dependent on lexical type and onset voicing. In function words (e.g., “across,” “only”), stressed syllables did not contain more symptoms than unstressed syllables, whereas in content words (e.g. “emu,” “salon”), stressed syllables did predict more symptoms. Notably, counter to our hypothesis, unstressed syllables in function words yielded more symptoms than those in content words (this result will be discussed further in the Lexical Type section below).

The effects of stress were also dependent on voicing contexts. Specifically, within voiceless consonant-initial syllables, stressed syllables were more likely to elicit AdLD symptoms. However, within vowel-initial syllables, the factor of stress was not found to be significant. This may be due to the influence of vowel–vowel contexts strongly eliciting symptom expression in speakers with AdLD (and eliciting glottalization in typical speakers, per Dilley et al., 1996), which could have dampened the effect of stress on these syllables. That is, vowel–vowel junctions lead to symptoms whether a syllable is stressed or not, but syllables with unvoiced onsets have greater symptoms when they are stressed. The effect of stress in unvoiced onset contexts may be due to speakers requiring more precise and/or effortful control to ensure their listener is perceiving their intended speech sounds. Typically, stressed syllables have increased vowel duration, increased intensity, and more extreme F0 and formant changes (Fry, 1958; Gay, 1978; Ladefoged, 1968; Van Summers, 1987). These acoustic cues require a greater magnitude of adjustments of articulatory structures (Van Summers, 1987) and increased subglottal pressure and laryngeal tension in speech production (Ladefoged, 1968). Other works indicate that lexical stress involves increased movement of the thyroarytenoid muscle specifically (Hirose & Gay, 1972), a muscle typically associated with symptoms in AdLD (Ludlow & Connor, 1987). More precise production when articulating stressed syllables may require increased effort and may exacerbate or trigger symptoms in speakers with AdLD.

Onset Voicing

In this experiment, we also found that syllables that began with a vowel had a greater probability of eliciting symptoms in AdLD speakers, and this was consistent across stress types and lexical types. This is in line with previous literature—increased symptom expression in speech with increased voiced sounds and vowel onsets in speakers with AdLD is well-documented clinically and experimentally (Cannito et al., 2014; Erickson, 2003; Simonyan et al., 2021). Additionally, the stimuli in this study consisted of vowel-initial syllables following a vowel-ending word, creating vowel–vowel syllable junctions in each voiced target syllable. Previous research suggests that typical speakers produce glottalization (e.g., glottal attack, creak) during those vowel–vowel word transitions (Dilley et al., 1996), consistent with the significant onset voicing effect (primarily the presence of creak) found in the present study's follow-up analysis in a control group of age- and sex-matched speakers. A previous study found that speakers with AdLD produced sound sequences that require glottal stops (/a/ repetitions) more slowly and with longer voicing gaps than control speakers but not sequences with alternating vowels and voiceless stops (/pa/ repetitions; Ludlow & Connor, 1987). Together with the present results, this suggests that they may have trouble with vowel–vowel transitions due to the tendency for English speakers to glottalize (and therefore, adduct the vocal folds) in these contexts. The exact mechanism of how glottalization may increase the frequency of symptoms in speakers with AdLD is unknown, although it has been suggested that motoric aspects of planning/executing these glottal movements (Reid & Nobriga, 2023) or changes in sensory feedback due to vocal fold adduction pressure or force may result in dystonic movements through central sensorimotor mechanisms (Izdebski, 1992). Future work should aim to examine these sensorimotor variables using other physiologic techniques such as electroglottography (as in Garellek, 2014) to better understand the precise mechanism.

These findings differ from those of Reid and Nobriga (2023), however, in that syllable-initial voicing did not have a significant effect on symptom expression. Reid and Nobriga discussed that syllable stress may have been confounded with voicing effects in their study since the majority of vowel-onset words in their stimuli were function words and hence were not stressed in reading. Additionally, the amount of glottalization associated with vowel onsets varies as a function of the preceding sound (Dilley et al., 1996) as well as other factors such as sentence position and/stress as demonstrated in previous work (e.g., Dilley et al., 1996; Garellek, 2014). Furthermore, this difference may be due to the different outcome measures used. While the outcome measure of the present study was the presence or absence of a feature of AdLD symptoms/glottalization based on visual inspection of the waveform, Reid and Nobriga used a measure of the consistency of symptom identification across multiple raters, with specific identification criteria not outlined. This difference in outcome measures between the two studies may have an unknown impact on the effect of onset voicing on symptom prevalence.

Lexical Type

In the present study, lexical type was not a significant linguistic factor overall in explaining AdLD symptoms when examined independent from other linguistic factors. Interestingly, however, the effect of lexical type was dependent on stress in an unexpected way—unstressed syllables in content words were less likely to contain symptoms than unstressed syllables in function words, but no effect of lexical type was found for stressed syllables. This was counter to our hypothesis and the results of previous studies (Froeschke, 2020; Reid & Nobriga, 2023). One way to explain these results relates to the fact that lexical stress is signaled by the contrast between unstressed and stressed syllables in a word (Mattys, 2000). As mentioned earlier, content words are centrally important to convey a message (Segalowitz & Lane, 2000) and thus may require greater contrast between stressed and unstressed syllables than in function words to be salient. This could be carried out by speakers either overemphasizing stressed syllables or deemphasizing unstressed syllables, similar to the concept of vowel reduction in unstressed syllables (Fourakis, 1991). This reduction in emphasis for unstressed syllables in content words may require less exertion and effort from the speaker and may thus result in fewer symptoms. Interestingly, results from our follow-up analyses with the control group also indicated similar results, with an effect of lexical type dependent on stress in eliciting AdLD symptoms. This may suggest that differences in symptoms across these linguistic factors are not specific to AdLD. It also suggests that AdLD may be more related to these linguistic factors as a result of their underlying motor actions rather than pure linguistic features of speech. Confirming these hypotheses will require direct testing in future work.

Our results may also differ from prior literature for methodological reasons. In both Froeschke (2020) and Reid and Nobriga (2023), symptoms were identified at the word level using purely auditory-perceptual methods, whereas the present study used specific acoustic guidelines related to the waveform of the signal (and visual inspection thereof). It is therefore possible that the more objective criteria of the present study led to more reliable symptom identification. Furthermore, Reid and Nobriga noted that, for their stimuli (which largely overlapped with those of Froeschke, the factor of syllable stress could not be separated from the factor of lexical type, since content words were often stressed single-syllable words in the stimuli. In the present study, when carefully controlling for the effects of lexical stress, the effect of lexical type was not significant. Finally, in the present study, only AdLD symptoms that appeared on the first syllable of two syllable words were retrieved for statistical analyses. In other words, all analysis was performed at the syllable level, whereas previous literature (Froeschke, 2020; Reid & Nobriga, 2023) performed analyses at the word or sentence level. It is possible that speakers in the present study had symptoms on the second syllables in these bisyllabic words, which were not captured by our analysis. This suggests that, in the present study, the number of symptoms found in content or function words may be decreased due to the analysis approach while examining the effect of lexical type and that the second syllable of bisyllabic words may have an important impact on the effect of lexical type.

Clinical Implications and Future Directions

Due to the rare nature of AdLD and its perceptual similarity to more common voice disorders, individuals with this disorder do not easily receive a diagnosis. The purpose of this study was to gain a better understanding of linguistic environments that may elicit symptoms at a higher probability in AdLD, which may aid in the design of better diagnostic tools for clinicians to use in the future. For example, using two sets of diagnostic speech sentences—one loaded with vowel–vowel word junctions, where content words are stressed, and another loaded with vowel–unvoiced consonant word junctions, where content words are unstressed—may elicit a maximal difference in symptom expression in speakers with AdLD. Alternatively, isolating individual words with these characteristics for perceptual or acoustic analysis may have a similar effect. At the same time, the diagnostic value of speech stimuli is critically dependent on how well they can discriminate AdLD from other voice disorders. Our secondary analysis on speakers with typical voices found similar word onset voicing contrasts and interactions between lexical type and stress, though with an overall decrease in AdLD voicing signs across all combinations of stimulus types (see Figure 2). This indicates that the significant effects of linguistic variables found herein for speakers with AdLD (including the vowel vs. voiceless consonant onset effects) may be found across English speakers in general, and thus, these contrasts may not be discriminative. Analysis of symptoms in a pMTD group and an AbLD group when reading the stimuli sentences would be critical to determine symptomatic differences between speakers with various voice disorders.

There were also several methodological choices that may need further examination in future work. An important voicing condition that was missing from the present investigation was voiced consonant onsets. In the AdLD symptom literature, vowels and voiced consonants are often lumped together when constructing voiced-loaded sentences (Erickson, 2003; Reid & Nobriga, 2023). While Cannito et al. (2014) found similar perceptual measures of severity across differing manners of voiced sounds at the sentence level, Frankford et al. (2025) showed significant distinctions when examining individual sounds. Therefore, examining symptoms expression on words beginning with these onsets may provide additional discriminatory information. There may also have been a ceiling effect of the symptoms elicited from vowel–vowel contexts observed in speakers with AdLD, whereby other effects of stress and lexical type were masked in this especially increased symptom environment. Analysis of symptoms elicited in more intermediate environments, such as vowel–voiced consonant contexts, may impact the effects of lexical type and lexical stress. Additionally, this study identified symptom expression in a binary manner, meaning that the presence of any symptom within a syllable made that syllable “symptomatic.” However, it may be useful to examine the presence or proportion of each symptom individually, as well as each symptom's prevalence in overlapping with other symptoms. For example, if one speaker with AdLD exhibited a large proportion of voice breaks, whereas another exhibited a majority of creaky voice segments, this may suggest different underlying mechanisms that are impacted by this dystonia, potential differences in the manifestation of this disorder, or separate subtypes of AdLD. Furthermore, several other linguistic factors may be observed and accounted for in more detail in future studies. For example, it may be beneficial to account for intended prosody patterns (pitch contours/intonation) in each sentence as well as sentence complexity (as in Erickson, 2003). Finally, even if the linguistic features in the present study do not provide clear discriminatory information for differential diagnosis, they may provide information on potential mechanisms of symptom expression and variability in AdLD. Future research will be necessary to understand the neurophysiological mechanisms, both central and peripheral, that lead to the variability of AdLD symptoms across contexts.

Conclusions

In the present study, a set of balanced speech stimuli was used to test the hypotheses that stressed syllables would elicit greater symptom expression than nonstressed syllables, vowel onsets would elicit greater symptom expression than voiceless consonants onsets, and content words would elicit greater symptom expression than function words in speakers with AdLD. The results of the study provide support for some of the hypotheses—syllable onset voicing and syllable stress, but not lexical type, were linguistic factors that significantly affected symptom expression in AdLD speakers overall. However, these effects were modulated by significant interactions between onset voicing and stress and between lexical type and stress. Although the overall effects of voicing and the Lexical Type × Stress interactions were also found for glottalization in control speakers, the Voicing × Stress interaction appeared to be specific to AdLD. These findings further our knowledge and understanding of the linguistic factors that contribute to symptom expression in AdLD. Understanding these linguistic factors has the potential to lead to improved diagnostic criteria for AdLD, provide clues for future work to examine the cause of task-specificity in AdLD, and ultimately improve pathways to effective treatment in individuals with AdLD.

Data Availability Statement

The data sets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.

Supplementary Material

Supplemental Material S1. Results of the primary analysis in speakers with AdLD without the Duration covariate.
JSLHR-69-4056-s001.pdf (579.8KB, pdf)
Supplemental Material S2. Estimated probability of symptom expression in speakers with AdLD across stress, voicing, and lexical type, without accounting for duration, reverse-transformed from the logit scale to the probability scale. Error bars indicate 95% confidence intervals.
JSLHR-69-4056-s002.pdf (592KB, pdf)
Supplemental Material S3. Results of the secondary analysis in control speakers.
JSLHR-69-4056-s003.pdf (576.9KB, pdf)
Supplemental Material S4. Estimated probability of symptom expression in the control group across stress, voicing, and lexical type, accounting for duration, reverse-transformed from the logit scale to the probability scale. Error bars indicate 95% confidence intervals.
JSLHR-69-4056-s004.pdf (529.3KB, pdf)
Supplemental Material S5. Estimated probability of symptom expression in the control group across lexical type and stress, collapsed across voicing conditions and accounting for duration. Estimates are reverse-transformed from the logit scale to the probability scale. Error bars indicate 95% confidence intervals.
JSLHR-69-4056-s005.pdf (511.8KB, pdf)

Acknowledgments

This work was supported by Grants R01 DC015570 (awarded to Cara E. Stepp) and F32 DC020349 (awarded to Katherine L. Marks) from the National Institute on Deafness and Other Communication Disorders. The authors wish to thank Daniel Buckley for his support in rating the severity of speakers and Jose Rojas for his help in recruiting participants.

Funding Statement

This work was supported by Grants R01 DC015570 (awarded to Cara E. Stepp) and F32 DC020349 (awarded to Katherine L. Marks) from the National Institute on Deafness and Other Communication Disorders.

References

  1. Blitzer, A., Brin, M. F., & Stewart, C. F. (1998). Botulinum toxin management of spasmodic dysphonia (laryngeal dystonia): A 12-year experience in more than 900 patients. The Laryngoscope, 108(10), 1435–1441. 10.1097/00005537-199810000-00003 [DOI] [PubMed] [Google Scholar]
  2. Bloch, C. S., Hirano, M., & Gould, W. J. (1985). Symptom improvement of spastic dysphonia in response to phonatory tasks. Annals of Otology, Rhinology & Laryngology, 94(1), 51–54. 10.1177/000348948509400111 [DOI] [PubMed] [Google Scholar]
  3. Boersma, P., & Weenink, D. (2019). Praat: Doing phonetics by computer (Version 6.1.03) [Computer software]. http://www.praat.org/
  4. Cannito, M. P., Chorna, L. B., Kahane, J. C., & Dworkin, J. P. (2014). Influence of consonant voicing characteristics on sentence production in abductor versus adductor spasmodic dysphonia. Journal of Voice, 28(3), 394.e13–394.e22. 10.1016/j.jvoice.2013.10.010 [DOI] [PubMed] [Google Scholar]
  5. Creighton, F. X., Hapner, E., Klein, A., Rosen, A., Jinnah, H. A., & Johns, M. M. (2015). Diagnostic delays in spasmodic dysphonia: A call for clinician education. Journal of Voice, 29(5), 592–594. 10.1016/j.jvoice.2013.10.022 [DOI] [PMC free article] [PubMed] [Google Scholar]
  6. Dilley, L., Shattuck-Hufnagel, S., & Ostendorf, M. (1996). Glottalization of word-initial vowels as a function of prosodic structure. Journal of Phonetics, 24(4), 423–444. 10.1006/jpho.1996.0023 [DOI] [Google Scholar]
  7. Edgar, J. D., Sapienza, C. M., Bidus, K., & Ludlow, C. L. (2001). Acoustic measures of symptoms in abductor spasmodic dysphonia. Journal of Voice, 15(3), 362–372. 10.1016/S0892-1997(01)00038-8 [DOI] [PubMed] [Google Scholar]
  8. Erickson, M. L. (2003). Effects of voicing and syntactic complexity on sign expression in adductor spasmodic dysphonia. American Journal of Speech-Language Pathology, 12(4), 416–424. 10.1044/1058-0360(2003/087) [DOI] [PubMed] [Google Scholar]
  9. Fourakis, M. (1991). Tempo, stress, and vowel reduction in American English. The Journal of the Acoustical Society of America, 90, 1816–1827. 10.1121/1.401662 [DOI] [PubMed] [Google Scholar]
  10. Frankford, S. A., Marks, K. L., Feaster, T. F., Doyle, P. C., & Stepp, C. E. (2025). Symptom expression across voiced speech sounds in adductor laryngeal dystonia. Journal of Voice, 39(2), 567.E23–567.E30. 10.1016/j.jvoice.2022.10.002 [DOI] [PMC free article] [PubMed] [Google Scholar]
  11. Froeschke, L. L. O. (2020). The influence of linguistic demand on symptom expression in adductor spasmodic dysphonia. Journal of Voice, 34(5), 807.e11–807.e21. 10.1016/j.jvoice.2019.04.003 [DOI] [PubMed] [Google Scholar]
  12. Fry, D. B. (1958). Experiments in the perception of stress. Language and Speech, 1(2), 126–152. 10.1177/002383095800100207 [DOI] [Google Scholar]
  13. Garellek, M. (2014). Voice quality strengthening and glottalization. Journal of Phonetics, 45(1), 106–113. 10.1016/J.WOCN.2014.04.001 [DOI] [Google Scholar]
  14. Gay, T. (1978). Effect of speaking rate on vowel formant movements. The Journal of the Acoustical Society of America, 63, 223–230. 10.1121/1.381717 [DOI] [PubMed] [Google Scholar]
  15. Hirose, H., & Gay, T. (1972). The activity of the intrinsic laryngeal muscles in voicing control. Phonetica, 25(3), 140–164. 10.1159/000259378 [DOI] [PubMed] [Google Scholar]
  16. Izdebski, K. (1992). Symptomatology of adductor spasmodic dysphonia: A physiologic model. Journal of Voice, 6(4), 306–319. 10.1016/S0892-1997(05)80027-X [DOI] [Google Scholar]
  17. Keating, P., Garellek, M., & Kreiman, J. (2015). Acoustic properties of different kinds of creaky voice. https://idiom.ucsd.edu/~mgarellek/files/Keating_etal_2015_ICPhS.pdf [PDF]
  18. Kempster, G. B., Gerratt, B. R., Verdolini Abbott, K., Barkmeier-Kraemer, J., & Hillman, R. E. (2009). Consensus auditory-perceptual evaluation of voice: Development of a standardized clinical protocol. American Journal of Speech-Language Pathology, 18(2), 124–132. 10.1044/1058-0360(2008/08-0017) [DOI] [PubMed] [Google Scholar]
  19. Ladefoged, P. (1968). Linguistic aspects of respiratory phenomena. Annals of the New York Academy of Sciences, 155(1), 141–151. 10.1111/j.1749-6632.1968.tb56758.x [DOI] [Google Scholar]
  20. Leonard, R., & Kendall, K. (1999). Differentiation of spasmodic and psychogenic dysphonias with phonoscopic evaluation. The Laryngoscope, 109(2), 295–300. 10.1097/00005537-199902000-00022 [DOI] [PubMed] [Google Scholar]
  21. Ludlow, C. L. (2011). Spasmodic dysphonia: A laryngeal control disorder specific to speech. Journal of Neuroscience, 31(3), 793–797. 10.1523/JNEUROSCI.2758-10.2011 [DOI] [PMC free article] [PubMed] [Google Scholar]
  22. Ludlow, C. L., Adler, C. H., Berke, G. S., Bielamowicz, S. A., Blitzer, A., Bressman, S. B., Hallett, M., Jinnah, H. A., Juergens, U., Martin, S. B., Perlmutter, J. S., Sapienza, C., Singleton, A., Tanner, C. M., & Woodson, G. E. (2008). Research priorities in spasmodic dysphonia. Otolaryngology—Head and Neck Surgery, 139(4), 495–505. 10.1016/j.otohns.2008.05.624 [DOI] [PMC free article] [PubMed] [Google Scholar]
  23. Ludlow, C. L., & Connor, N. P. (1987). Dynamic aspects of phonatory control in spasmodic dysphonia. Journal of Speech and Hearing Research, 30(2), 197–206. 10.1044/jshr.3002.197 [DOI] [PubMed] [Google Scholar]
  24. Ludlow, C. L., Domangue, R., Sharma, D., Jinnah, H. A., Perlmutter, J. S., Berke, G., Sapienza, C., Smith, M. E., Blumin, J. H., Kalata, C. E., Blindauer, K., Johns, M., Hapner, E., Harmon, A., Paniello, R., Adler, C. H., Crujido, L., Lott, D. G., Bansberg, S. F., … Stebbins, G. (2018). Consensus-based attributes for identifying patients with spasmodic dysphonia and other voice disorders. JAMA Otolaryngology—Head and Neck Surgery, 144(8), 657–665. 10.1001/jamaoto.2018.0644 [DOI] [PMC free article] [PubMed] [Google Scholar]
  25. Ludlow, C. L., Naunton, R. F., Sectary, S. E., Schulz, G. M., & Hallett, M. (1988). Effects of botulinum toxin injections on speech in adductor spasmodic dysphonia. Neurology, 38(8), Article 1220. 10.1212/WNL.38.8.1220 [DOI] [PubMed] [Google Scholar]
  26. Marks, K. L., Frankford, S. A., Cocroft, S. J., Lonergan, S., Díaz Cádiz, M. E., & Stepp, C. E. (2024). The prevalence of creak across breath groups in adductor laryngeal dystonia. Journal of Voice. Advance online publication. 10.1016/j.jvoice.2024.10.013 [DOI] [PMC free article] [PubMed] [Google Scholar]
  27. Marks, K. L., Feaster, T. F., Baker, S., Díaz-Cádiz, M. E., Doyle, P. C., & Stepp, C. E. (2022). Spectral aggregate of the high-passed fundamental frequency and its relationship to the primary acoustic features of adductor laryngeal dystonia. Journal of Speech, Language, and Hearing Research, 65(11), 4085–4095. 10.1044/2022_JSLHR-22-00157 [DOI] [PMC free article] [PubMed] [Google Scholar]
  28. Mattys, S. L. (2000). The perception of primary and secondary stress in English. Perception & Psychophysics, 62(2), 253–265. 10.3758/BF03205547 [DOI] [PubMed] [Google Scholar]
  29. McGraw, K. O., & Wong, S. P. (1996). Forming inferences about some intraclass correlation coefficients. Psychological Methods, 1(1), 30. https://psycnet.apa.org/doi/10.1037/1082-989X.1.1.30 [Google Scholar]
  30. Murry, T., & Woodson, G. E. (1992). A comparison of three methods for the management of vocal fold nodules. Journal of Voice, 6(3), 271–276. 10.1016/S0892-1997(05)80153-5 [DOI] [Google Scholar]
  31. Reid, E. W., & Nobriga, C. V. (2023). Spasmodic dysphonia: Introductory phonetic analyses. Clinical Linguistics & Phonetics, 37(10), 883–898. 10.1080/02699206.2022.2096483?needAccess=true [DOI] [PubMed] [Google Scholar]
  32. Roy, N., Awan, S. N., Jennings, S., Jensen, J., & Merrill, R. M. (2024). Adductor laryngeal dystonia versus muscle tension dysphonia: Examining the utility of automated acoustic analysis to detect task dependency as a distinguishing feature. Journal of Speech, Language, and Hearing Research, 67(10), 3612–3630. 10.1044/2024_JSLHR-24-00104 [DOI] [PubMed] [Google Scholar]
  33. Roy, N., Barkmeier-Kraemer, J., Eadie, T., Sivasankar, M. P., Mehta, D., Paul, D., & Hillman, R. (2013). Evidence-based clinical voice assessment: A systematic review. American Journal of Speech-Language Pathology, 22(2), 212–226. 10.1044/1058-0360(2012/12-0014) [DOI] [PubMed] [Google Scholar]
  34. Roy, N., Mauszycki, S. C., Merrill, R. M., Gouse, M., & Smith, M. E. (2007). Toward improved differential diagnosis of adductor spasmodic dysphonia and muscle tension dysphonia. Folia Phoniatrica et Logopaedica, 59(2), 83–90. 10.1159/000098341 [DOI] [PubMed] [Google Scholar]
  35. Sapienza, C. M., Cannito, M. P., Murry, T., Branski, R., & Woodson, G. (2002). Acoustic variations in reading produced by speakers with spasmodic dysphonia pre-Botox injection and within early stages of post-Botox injection. Journal of Speech, Language, and Hearing Research, 45(5), 830–843. 10.1044/1092-4388(2002/067) [DOI] [PubMed] [Google Scholar]
  36. Sapienza, C. M., Walton, S., & Murry, T. (1999). Acoustic variations in adductor spasmodic dysphonia as a function of speech task. Journal of Speech, Language, and Hearing Research, 42(1), 127–140. 10.1044/jslhr.4201.127 [DOI] [PubMed] [Google Scholar]
  37. Segalowitz, S. J., & Lane, K. C. (2000). Lexical access of function versus content words. Brain and Language, 75(3), 376–389. 10.1006/BRLN.2000.2361 [DOI] [PubMed] [Google Scholar]
  38. Silverman, E. P., Garvan, C., Shrivastav, R., & Sapienza, C. M. (2012). Combined modality treatment of adductor spasmodic dysphonia. Journal of Voice, 26(1), 77–86. 10.1016/J.JVOICE.2010.08.004 [DOI] [PubMed] [Google Scholar]
  39. Simonyan, K., Barkmeier-Kraemer, J., Blitzer, A., Hallett, M., Houde, J. F., Jacobson Kimberley, T., Ozelius, L. J., Pitman, M. J., Richardson, R. M., Sharma, N., Tanner, K., & on behalf of the The NIH/NIDCD Workshop on Research Priorities in Spasmodic Dysphonia/Laryngeal Dystonia . (2021). Laryngeal dystonia: Multidisciplinary update on terminology, pathophysiology, and research priorities. Neurology, 96(21), 989–1001. 10.1212/WNL.0000000000011922 [DOI] [PMC free article] [PubMed] [Google Scholar]
  40. Simonyan, K., & Ludlow, C. L. (2012). Abnormal structure–function relationship in spasmodic dysphonia. Cerebral Cortex, 22(2), 417–425. 10.1093/cercor/bhr120 [DOI] [PMC free article] [PubMed] [Google Scholar]
  41. Simonyan, K., Ostuni, J., Ludlow, C. L., & Horwitz, B. (2009). Functional but not structural networks of the human laryngeal motor cortex show left hemispheric lateralization during syllable but not breathing production. Journal of Neuroscience, 29(47), 14912–14923. 10.1523/JNEUROSCI.4897-09.2009 [DOI] [PMC free article] [PubMed] [Google Scholar]
  42. Van Summers, W. (1987). Effects of stress and final-consonant voicing on vowel production: Articulatory and acoustic analyses. The Journal of the Acoustical Society of America, 82, 847–863. 10.1121/1.395284 [DOI] [PubMed] [Google Scholar]
  43. Whurr, R., & Lorch, M. (2016). Review of differential diagnosis and management of spasmodic dysphonia. Current Opinion in Otolaryngology & Head and Neck Surgery, 24(3), 203–207. 10.1097/MOO.0000000000000253 [DOI] [PubMed] [Google Scholar]
  44. Wolraich, D., Vasile Marchis-Crisan, C., Redding, N., Khella, S. L., & Mirza, N. (2010). Laryngeal tremor: Co-occurrence with other movement disorders. ORL, 72(5), 291–294. 10.1159/000317032 [DOI] [PubMed] [Google Scholar]
  45. Yuan, J., & Liberman, M. (2008). Speaker identification on the SCOTUS corpus. The Journal of the Acoustical Society of America, 123, Article 3878. 10.1121/1.2935783 [DOI] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Supplemental Material S1. Results of the primary analysis in speakers with AdLD without the Duration covariate.
JSLHR-69-4056-s001.pdf (579.8KB, pdf)
Supplemental Material S2. Estimated probability of symptom expression in speakers with AdLD across stress, voicing, and lexical type, without accounting for duration, reverse-transformed from the logit scale to the probability scale. Error bars indicate 95% confidence intervals.
JSLHR-69-4056-s002.pdf (592KB, pdf)
Supplemental Material S3. Results of the secondary analysis in control speakers.
JSLHR-69-4056-s003.pdf (576.9KB, pdf)
Supplemental Material S4. Estimated probability of symptom expression in the control group across stress, voicing, and lexical type, accounting for duration, reverse-transformed from the logit scale to the probability scale. Error bars indicate 95% confidence intervals.
JSLHR-69-4056-s004.pdf (529.3KB, pdf)
Supplemental Material S5. Estimated probability of symptom expression in the control group across lexical type and stress, collapsed across voicing conditions and accounting for duration. Estimates are reverse-transformed from the logit scale to the probability scale. Error bars indicate 95% confidence intervals.
JSLHR-69-4056-s005.pdf (511.8KB, pdf)

Data Availability Statement

The data sets generated during and/or analyzed during the current study are available from the corresponding author on reasonable request.

Articles from Journal of Speech, Language, and Hearing Research : JSLHR are provided here courtesy of American Speech-Language-Hearing Association

RESOURCES