The Contributions of Pitch, Loudness, and Rate Control to Speech Naturalness in Cerebellar Ataxia

Abstract

Purpose:

The goal of this study was to determine the relationship between the perceptual measure of speech naturalness and objective measures of pitch, loudness, and rate control as a potential tool for assessment of ataxic dysarthria.

Method:

Twenty-seven participants with ataxia and 29 age- and sex-matched control participants completed the pitch glide and loudness step tasks drawn from the Frenchay Dysarthria Assessment–Second Edition (FDA-2) in addition to speech diadochokinetic (DDK) tasks. First, group differences were compared for pitch variability in the pitch glide task, loudness variability in the loudness step task, and syllable duration and speech rate in the DDK task. Then, these acoustic measures were compared with previously collected ratings of speech naturalness by speech-language pathology graduate students.

Results:

Robust group differences were measured for pitch variability and both DDK syllable duration and speech rate, indicating that the ataxia group had greater pitch variability, longer DDK syllable duration, and slower DDK speech rate than the control group. No group differences were measured for loudness variability. There were robust relationships between speech naturalness and pitch variability, DDK syllable duration, and DDK speech rate, but not for loudness variability.

Conclusions:

Objective acoustic measures of pitch variability in the FDA-2 pitch glide task and syllable duration and speech rate in the DDK task can be used to validate perceptual measures of speech naturalness. Overall, speech-language pathologists can incorporate both perceptual measures of speech naturalness and acoustic measures of pitch variability and DDK performance for a comprehensive evaluation of ataxic dysarthria.

Assessment of ataxic dysarthria presents challenges to speech-language pathologists (SLPs), because intelligibility is often minimally impacted regardless of severity (Hilger et al., 2022). Ataxic dysarthria is a neuromotor speech disorder typically caused by cerebellar dysfunction, characterized by aberrant production of prosody, articulation, and vocal quality (Kent et al., 2000). The speech impairments in ataxia typically manifest as reduced speech naturalness and prosodic abnormality (Hilger et al., 2022; Kent et al., 2000), each of which must be assessed perceptually due to a lack of validated objective assessment measures for clinical use. Although the acoustic characteristics of ataxic dysarthria have been characterized, translation of these findings for objective clinical assessment is limited. The goal of this study is to evaluate the relationship between auditory-perceptual ratings of speech naturalness and objective acoustic measures of pitch, loudness, and rate control in simple production tasks to determine whether these measures may serve as possible tools for objective assessment of ataxic dysarthria.

The distinct profile of prosodic features that is characteristic of ataxic dysarthria may have a substantial impact on quality of life, yet the underlying mechanisms that lead to disrupted prosody have yet to be fully explored. This gap in current knowledge hinders the development of effective speech interventions for individuals with ataxia. The neurological mechanisms and acoustic correlates that lead to perceptual speech changes in ataxia must be elucidated for clinical research to progress and for optimal interventions to be established. This study attempts to bridge this gap by determining which acoustic features strongly relate to our perception of reduced naturalness, which could help improve our knowledge of underlying physiological mechanisms contributing to prosodic impairment.

Overview of Cerebellar Ataxia

Cerebellar ataxia is a clinical sign that occurs in a cluster of heterogeneous, debilitating neurodegenerative diseases as a result of damage to the cerebellum and/or its pathways (Beaudin et al., 2019; de Silva et al., 2019; Sidtis et al., 2011). Ataxia is characterized by deficits in motor control, which affect the coordination and precision of movements (Ackermann, 2008; Folker et al., 2010; Kent et al., 1979). Classic signs and symptoms of cerebellar ataxia include difficulties with voluntary limb movements, balance and gait dysfunction, oculomotor disturbances, and dysarthria (Manto et al., 2012; Marsden & Harris, 2011).

The cerebellum is involved in combining learned movements together with precise timing to produce skilled motor behavior (Itō, 1984; Ivry et al., 2002; Laforce & Doyon, 2001). Different internal models of cerebellar motor prediction and control have been proposed, but the precise mechanisms of these functions remain under investigation (Manto et al., 2012). Nonetheless, a consensus has emerged that the cerebellum is involved in generating predictions of the sensory consequences of movement (Blakemore et al., 2001; Knolle et al., 2012). The cerebellum then integrates sensory feedback with cortical motor plans to adjust movement parameters as needed based on specific contextual demands (Ackermann, 2008).

Speech is a complex motor behavior that is highly variable across contexts (Schulz et al., 2005). Rapid, precise, coordinated movement of oral, velar, laryngeal, and respiratory muscles is required to produce speech that is both intelligible and natural sounding (Yorkston & Beukelman, 1981). Although numerous neural regions are implicated in the programming and execution of movement, the cerebellum is fundamentally involved in the precise sequencing and control of the movements that lead to speech production (Mariën et al., 2014; Spencer & Rogers, 2005). As such, the disruption of cerebellar circuits in ataxia often impacts speech. Changes in speech may emerge as one of the early clinical signs of ataxia in some patients (Bodranghien et al., 2016), yet patients may not always be referred for speech evaluation and intervention given the limited understanding of ataxia symptomatology and treatment across disciplines (Daker-White et al., 2013; de Silva et al., 2019; Hilger & Dunne-Platero, 2022).

Characteristics of Ataxic Dysarthria

Speech characteristics of ataxic dysarthria have been classified into three distinct clusters: articulatory inaccuracy, prosodic excess, and phonatory-prosodic insufficiency (Darley et al., 1969; Duffy, 2019). Articulatory inaccuracy encompasses features such as irregular articulatory breakdowns, imprecise consonants, vowel distortions, and telescoping of syllables (Darley et al., 1969; Duffy, 2019; Kent et al., 1979). Prosodic excess includes characteristics such as excess and equal stress, prolonged phonemes, prolonged interword intervals, and slow rate of speech (Darley et al., 1969), which may give speech a “scanning” quality or a word-by-word cadence (Hartelius et al., 2000; Kent & Rosenbek, 1982). Finally, phonatory-prosodic insufficiency refers to characteristics such as harsh vocal quality, monopitch, monoloudness, and occasional voice stoppage (Ackermann & Hertrich, 2000; Darley et al., 1969; Joanette & Dudley, 1980).

The clinical profile of impairments that presents in ataxic dysarthria varies depending on the etiology and progression of cerebellar pathology. Thus, significant heterogeneity exists in the clinical presentation of “pure” ataxic dysarthria (Spencer & Dawson, 2019; Spencer & France, 2016). Furthermore, the progression of degenerative hereditary ataxias over time can impact other areas of the nervous system, which may lead to the development of mixed dysarthria (Folker et al., 2010; Poole et al., 2015; Taroni & DiDonato, 2004).

Notably, intelligibility is often relatively spared in ataxic dysarthria (Blaney & Hewlett, 2007; Brendel et al., 2013; Hilger et al., 2022). Intelligibility can be defined as the degree to which a listener understands the acoustic signal produced by a speaker (Kent et al., 1989; Yorkston et al., 1996). However, individuals with ataxia experience significant changes in the naturalness and efficiency of their speech, reflected by their prosodic impairment (Hilger et al., 2022; Yorkston & Beukelman, 1981). Naturalness is a perceptually derived measure that can be defined as the degree to which a listener judges a speech sample as following typical standards of rate, rhythm, intonation, and stress patterning (Yorkston et al., 1996). Even if intelligibility is preserved, impairments in speech naturalness can significantly disrupt quality of life. Patients often report a “slurred” or “drunken” quality of speech and frequent stumbling over words (Duffy, 2019). These qualities may ultimately lead to reduced life participation in patients with ataxic dysarthria due to the negative reactions of listeners, changes in self-perception, and environmental barriers to communication (Hartelius et al., 2007; Hilger & Dunne-Platero, 2022; Walshe & Miller, 2011).

Challenges to Perceptual Assessment of Ataxic Dysarthria

There are many challenges to perceptual assessment of speech in ataxia. First, there are inherent challenges in auditory-perceptual evaluation of motor speech disorders in general, which include clinician bias related to expertise and overlap of deviant speech features among dysarthria subtypes (Pernon et al., 2022). SLPs who frequently encounter patients with ataxia may be more familiar with patterns of deviant speech characteristics in ataxic dysarthria and may have stronger perceptual skills compared to less-experienced clinicians. Furthermore, prosody and speech naturalness present challenges for perceptual assessment compared to other speech features such as intelligibility. Intelligibility can be assessed with relative objectivity by identifying the percentage of words that are understood correctly in a speech sample, although this measurement can be influenced by factors such as listener familiarity with the patient's speech and the listening conditions (Yorkston & Beukelman, 1978; Yorkston et al., 1996). Prosody and speech naturalness, on the other hand, do not lend themselves to similar objective measures. Although we know that the perception of pitch, loudness, and timing influences the judgment of prosody and naturalness, there is so much variability in the production of natural-sounding prosody that it can be difficult to evaluate what constitutes impaired prosody and which features are contributing to the perception of reduced naturalness (Cole, 2015). Establishing objective acoustic measures to validate perceptual ratings would be beneficial for improving reliability in assessment.

Investigators have begun to explore the acoustic correlates of perceptual speech judgments in populations with ataxia. Perceptual judgments of atypical pitch and loudness in ataxic dysarthria have been characterized acoustically by features such as phonatory instability (Ackermann & Ziegler, 1994; Boutsen et al., 2010), reduced variability in fundamental frequency (F0; Casper et al., 2007), exaggerated F0 movements (Kent & Rosenbek, 1982), and either reduced or excess variation in intensity (Boutsen et al., 2010; Lowit et al., 2010; Spencer & France, 2016). The prosodic impairment in ataxia has further been characterized by features such as increased numbers of pitch accents produced per phrase (Lowit & Kuschmann, 2012), increased pause lengths (Rosen et al., 2003), atypical speech rate and syllabic timing (Ackermann & Hertrich, 1994), atypical and inconsistent assignment of sentence stress (Lowit et al., 2014), and impaired auditory feedback control (Hilger, 2020). Articulatory incoordination in ataxia has been characterized by features such as coarticulatory errors (Hertrich & Ackermann, 1999) and slower syllable durations in sentence production (Reilly & Spencer, 2013; Ziegler & Wessel, 1996).

Current Study

Although these articles have begun to describe impairments in prosody, pitch, loudness, and articulation in populations with ataxia, there is limited evidence describing how pitch, loudness, and rate control specifically relate to the perception of impaired prosody and reduced speech naturalness. The aim of the present study is to examine the acoustic correlates of pitch, loudness, and rate control and their relationship to speech naturalness in speakers with ataxia as compared to neurologically healthy control speakers. Specifically, we examine the variability in production during a pitch glide task, loudness step task, and diadochokinetic (DDK) rate task. DDK rate is a measure of articulatory coordination and sequencing that is sensitive to the coordinative impairment from cerebellar damage in ataxia (Kent et al., 1997). For this task, patients are prompted to first repeat single syllables of “puh-puh-puh,” “tuh-tuh-tuh,” and “kuh-kuh-kuh” and then repeat a sequence of these syllables of “puh-tuh-kuh.” Both the repeated syllable task (termed alternating motion rate [AMR]) and the sequential syllable task (termed sequential motion rate [SMR]) require considerable coordination for accurate production. First, laryngeal adjustment is required to switch between the unvoiced consonants /p, t, k/ and the voiced vowels (Kent et al., 2021). Next, precise timing is required for the rapid production of repeated syllables during the AMR task, which is challenging in ataxia for repeated speech and nonspeech tasks (Nguyen et al., 2020). The speaker must time the articulatory, respiratory, and phonatory movements so that there is a steady and rapid rhythm of repeated syllables. Because movement timing is impaired from cerebellar dysfunction, there is typically an irregular AMR rhythm in ataxic dysarthria (Kent et al., 2000). Furthermore, both the AMR and SMR tasks require a faster rate than is produced in everyday speech, adding an even greater coordinative demand on the speech musculature (Rong & Heidrick, 2021). Finally, to accurately produce the SMR task, the patient must coordinate lip and tongue movements to switch between the bilabial /p/, alveolar /t/, and velar /k/ while continuing to coordinate respiratory and phonatory systems for the voiceless consonants and voiced vowels (Kent et al., 2021). The goal of the task is to produce these three syllables as rapidly and consistently as possible despite the challenge of differences in articulatory speed and control for the lips (/p/), tongue tip (/t/), and tongue dorsum (/k/), with the tongue dorsum being the slowest articulator of the three (Ostry & Munhall, 1985). Overall, the DDK task, despite its simplicity, is effective in highlighting coordinative deficits.

We hypothesize that speech coordination impairments in cerebellar ataxia, which impact prosody and speech naturalness, are reflected in the objective acoustic information that can be gathered during simple commonly used speech production tasks, such as pitch glide, loudness step, and DDK tasks. Although the acoustic characterization of impaired prosody is not novel and has been extensively studied, an analysis determining the relationship between these characteristics and the perception of speech naturalness is novel and could be highly beneficial for clinical assessment. Furthermore, the measurement of acoustic features in tasks that are simple to administer (and are already commonly used) further demonstrates the potential for clinical translation of this work. This finding of a relationship between acoustic data and perceptual naturalness ratings would allow clinicians to compare their perceptual ratings with objective measures for alignment. Additionally, the objective measurement of these speech tasks has potential use for automatized assessment of speech naturalness in the future. In line with our hypothesis, we predict that there will be greater variability in pitch, loudness, and rate control in the ataxia group compared to neurologically healthy control participants. Furthermore, we predict that greater variability in all three of these acoustic measures will correlate with lower ratings of speech naturalness.

Method

Participants

Participants With Ataxia

The data analyzed in this study are from a data set of participants from Hilger et al.'s (2022) study. A brief summary of participant characteristics will be provided here, but more detail can be found in the original article. A total of 27 participants with ataxia were recruited (nine men, 18 women; 24–79 years of age; M = 54.3, SD = 15.1). Education ranged from 12 to 22 years (M = 15.3, SD = 2.5). All participants were native speakers of American English. Participants had normal, or corrected to normal, visual acuity. Ataxia diagnosis was confirmed through participant self-reports of neurology and/or genetic testing. Participants were recruited through local support groups, outpatient clinics of local medical/rehabilitation facilities, flyers in the monthly National Ataxia Foundation newsletter (National Ataxia Foundation, 2016), social media, word of mouth, the Communication Research Registry at Northwestern University, and the Coordination of Rare Diseases at Sanford registry (Trudeau, 2013). Summary characteristics of speakers with ataxia are provided in Table 1. Dysarthria type and severity were assessed using the Frenchay Dysarthria Assessment–Second Edition (FDA-2; Enderby & Palmer, 2008). For dysarthria type, the FDA-2 scoring form provides a pseudo-density plot of the observed deficits to compare with pseudo-density plots in the assessment manual by dysarthria subtype. For example, in individuals with ataxic dysarthria, normal function is expected for reflexes and respiration whereas more abnormal function is expected in lip, jaw, and tongue movement tasks. The resulting dysarthria subtype was diagnosed by comparing the level of function across the domains in the FDA-2 with the expected levels by subtype in the manual (Eigentler et al., 2012). All participants had a primary diagnosis of ataxic dysarthria. The FDA-2 does not have a method for diagnosing mixed dysarthria. Although mixed dysarthria profiles of ataxic-spastic or ataxic-flaccid are possible in cerebellar disease, mixed types were not assigned to participants in this study; instead, the primary single type was used as a descriptor. Dysarthria severity was similarly assessed by comparing the level of function across the tested speech subsystems. A mild dysarthria was diagnosed if the functioning level across most of the subsystems was close to normal or a little lower than normal, a moderate dysarthria if functioning was more in the middle range, and a severe dysarthria if functioning was closer to the low end of the range across subsystems. Mild–moderate and moderate–severe severity were diagnosed if there was some variation between these three levels. This method of assessing severity uses a combination of clinical judgment (rather than using an objective measure like speaking rate) based on the level of functioning across speech subsystems observed in the FDA-2. Assessment of severity of dysarthric speech using clinical judgment has been found to be both reliable and valid (Stipancic et al., 2021). This study was approved by the Northwestern University Institutional Review Board.

Table 1.

Participant characteristics.

Participant group	Participant number	Sex	Age	Education (years)	Ataxia diagnosis	Disease duration (years)	Dysarthria severity
AT	1	M	67	14	SCA-unknown	2.5	Mild
AT	2	M	47	14	SCA-unknown	23	Mild–moderate
AT	3	M	72	22	SCA6	3	Severe
AT	4	F	62	14	SCA6	1	Mild–moderate
AT	5	F	42	16	SCA2	0.5	Mild–moderate
AT	6	M	36	12	SCA7	0.5	Mild
AT	7	M	55	14	SCA1	22	Severe
AT	8	M	24	14	SCA2	3	Mild
AT	9	F	67	16	SCA6	20	Mild–moderate
AT	10	F	41	18	SCA3	10	Mild–moderate
AT	11	F	55	14	SCA3	0.5	Mild
AT	12	F	63	14	SCA6	3	Mild
AT	13	F	69	15	SCA-unknown	10	Moderate
AT	14	F	70	16	SCA3	5	Mild
AT	15	M	64	12	SCA15	24	Mild
AT	16	F	65	14	SCA-unknown	7	Mild–moderate
AT	17	F	62	18	Gluten ataxia	14	Mild
AT	18	F	36	18	SCA5	13	Mild
AT	19	F	42	18	AOA2	23	Mild–moderate
AT	20	F	60	18	SCAR8	21	Mild–moderate
AT	21	M	55	16	FA	14	Mild–moderate
AT	22	F	76	14	SCA6	9	Mild–moderate
AT	23	F	55	18	SCA-unknown	2	Moderate
AT	24	F	79	12	SCA-unknown	3	Mild
AT	25	M	31	12	FA	0.5	Mild–moderate
AT	26	F	47	18	SCA-unknown	25	Mild–moderate
AT	27	F	28	12	FA	12	Mild–moderate
CO	1	M	68	18
CO	2	M	45	16
CO	3	M	71	18
CO	4	F	61	12
CO	5	F	38	18
CO	6	M	38	18
CO	7	M	55	18
CO	8	M	24	16
CO	9	F	65	16
CO	10	F	40	16
CO	11	F	51	12
CO	12	F	66	18
CO	13	F	70	22
CO	14	F	70	18
CO	15	M	63	18
CO	16	F	63	18
CO	17	F	60	18
CO	18	F	36	22
CO	19	F	41	18
CO	20	F	58	18
CO	21	M	50	18
CO	22	F	71	16
CO	23	F	79	16
CO	24	F	54	18
CO	25	M	36	18
CO	26	F	42	18
CO	27	F	23	20
CO	28	F	62	18
CO	29	M	70	18

Note. Participants are listed by group (AT = ataxia; CO = control), participant number, sex (M = male; F = female), education, ataxia diagnosis (SCA = spinocerebellar ataxia; AOA = ataxia with oculomotor apraxia; SCAR = spinocerebellar ataxia recessive autosomal; FA = Friedreich's ataxia), disease duration, and dysarthria severity.

Healthy Control Speakers

Twenty-nine adults, with no reported history of speech, language, or neurological impairment, were recruited for this study as age- and sex-matched control participants (10 men, 19 women). All participants were native speakers of American English. Ages ranged from 24 to 79 years (M = 54.1, SD = 15.0). Years of education ranged from 12 to 22 years (M = 17.3, SD = 2.1). Participants had normal, or corrected to normal, visual acuity. Participants were recruited through social media and word of mouth.

Experiment Overview

Transportation was limited for many of the study participants with ataxia, so participants were provided with four options for testing sites: (a) the Speech Physiology Lab at Northwestern University, (b) the Neurology Clinic at the Northwestern Memorial Hospital, (c) a rented office space in Downtown Chicago, or (d) in a quiet room in their home. This experiment is part of a larger experiment examining auditory feedback control in ataxia. For the purpose of this study, only the following tasks were used: the pitch glide task from the FDA-2 to assess pitch control, the loudness step task from the FDA-2 to assess loudness control, DDK tasks to assess articulatory coordination, and speech samples from a picture description task (Cookie Theft Picture) and a spontaneous speech task (prompt to talk about a typical day) to gather naturalness ratings.

For the pitch glide task in the FDA-2, the participant is instructed to sing up and down a scale of seven notes of their choosing. For the loudness step task in the FDA-2, the participant is instructed to count from one to five with increasing loudness on each number. These tasks were administered according to the directions in the FDA-2 by the SLP (author A.H.). To ensure task compliance, the clinician monitored the pitch glide and loudness step tasks during the testing sessions, and if a participant made an error, the protocol was to have the participant repeat the task. An error would be defined as singing too few notes in the pitch glide task, or saying the wrong number when doing the loudness step task (e.g., saying, “One, three, four, six, seven,” as an example, instead of counting, “One, two, three, four, five”). However, the clinician did not observe any errors in the production of these two tasks. The DDK task was administered as typical with instructions to repeat the alternating syllables of /pʌpʌpʌ/ (“puhpuhpuh”), /tʌtʌtʌ/ (“tuhtuhtuh”), and /kʌkʌkʌ/ (“kuhkuhkuh”) and the sequential syllables of /pʌtʌkʌhpʌtʌkʌh/ (“puhtuhkuhpuhtuhkuh”) as quickly and as evenly as possible.

Instrumentation

Participants vocalized into an over-ear microphone (AKG, model C 420) positioned approximately 1 in. from the corner of the mouth. The microphone signal was digitized with a MOTU UltraLite-mk3 and recorded in a multichannel recording system (AD Instruments, model ML785, PowerLab A/D converter) and LabChart software (AD Instruments, v.7.0) with a sampling rate of 20 kHz.

Analysis

Naturalness ratings

Naturalness ratings were obtained from a prior study by Hilger et al. (2022). As a summary, 10 speech-language pathology graduate students listened to four randomly selected phrases per participant (two from picture description and two from spontaneous speech) and rated the naturalness of each sample on an equal-appearing interval (EAI) scale of 1–7, with 1 indicating a profound impairment in naturalness and 7 indicating no impairment. Speech naturalness was defined for the raters as how well the sample adhered to the rater's standard of rate, rhythm, intonation, and stress patterning. The phrases were chosen by exporting the transcribed utterances into a spreadsheet and using a random number generator to select two phrases per task per participant. As described by Hilger et al. (2022), each audio sample was 5–7 s in length, and there was no statistically significant difference in the phonetic complexity of the phrases among the samples. As a short summary, phonetic complexity was calculated using a modified consonant classification system by Allison and Hustad (2014) modified from Kim et al. (2010) in which consonants from each utterance were categorized into five levels of articulatory complexity. The resulting complexity score per phrase is a sum of these values. The nonstatistical difference for phonetic complexity between participant groups, t(218) = −1.29, p = .19, d = 0.17, demonstrates that participants with ataxia did not use more simple articulatory movement than control speakers, which could affect perceptual estimates.

To rate naturalness, an EAI scale was utilized for ease of use given that speech naturalness has been conceptualized as a metathetic dimension of speech, which lends itself to valid measurement via either interval scaling or direct magnitude estimation (Eadie & Doyle, 2002; Metz et al., 1990) and that speech naturalness has most commonly been measured using EAI scales in the speech-language pathology literature (Klopfenstein et al., 2020). For measures of rater reliability, 15% of the trials were duplicated. For statistical measures of reliability, the “irr” function from the R package “psych” was used to calculate intraclass correlation coefficients (ICCs; Revelle & Revelle, 2015). A two-way ICC analysis of consistency among raters using an “average unit” was implemented according to guidelines by Koo and Li (2016). High inter- and intrarater reliability was measured. The ICC score for intrarater reliability of naturalness was .92, indicating high reliability (95% confidence interval [0.90, 0.94], p < .0001). The ICC score for interrater reliability of naturalness was .96, indicating high reliability (95% confidence interval [0.94, 0.97], p < .0001). For this study, the naturalness scores for each speaker were averaged per listener and then averaged across listeners so that each speaker had one naturalness score.

Acoustic Analysis

The goal of the acoustic analysis was to obtain measures of variability for each task: the pitch glide task (further called the “pitch task”), the loudness step task (further called the “loudness task”), and the DDK task. For the pitch task, the mean F0 of the midline of each sung note was measured in Praat (Boersma & Weenink, 2019). Manual inspection determined that the F0 value was not erroneous (i.e., due to pitch tracking errors in Praat). These measurements were then converted from hertz to cents to standardize the values across participants using the following formula: cents = 1,200(log₂[f2/f1]), where f1 equals the mean F0 from a sustained phonation task in which participants held an “ahh” vowel for 3 s at a comfortable pitch and loudness and f2 equals the mean F0 of the midline of each note. We then calculated the difference in pitch (in cents) of each step in pitch from note to succeeding note, averaged these pitch steps, and calculated the coefficient of variance by dividing the standard deviation of the pitch steps over the average of the pitch steps. The coefficient of variance for pitch was our final dependent variable as an indicator of variability in pitch production.

For the loudness task, the mean intensity was measured for each number spoken. We then calculated the difference in intensity (dB) of each step in intensity from one number to the next number, averaged these intensity steps, and calculated the coefficient of variance by dividing the standard deviation of the intensity steps over the average of the intensity steps. The coefficient of variance for intensity was our final dependent variable as an indicator of variability in loudness production.

For the DDK analysis, we segmented and transcribed the phonemes in the DDK tasks using Praat. For example, the consonants /p/, /t/, and /k/ were segmented from the vowel /ə/ using the spectrogram as a guide. The vowels were defined as the points in the spectrogram where clear formants could be seen. The onset of the consonants (all voiceless stops) was defined as the onset of the silent phase of the stop (visualized as a short period of silence in the spectrogram) and the offset of the consonants as the offset of the release burst (visualized as a short column of noise in the spectrogram). After the transcription and segmentation processes, custom Praat scripts were used to extract measures of syllable duration (in seconds) and production rate (syllables/second) for each participant in each trial. For syllable duration, only the duration of the vowel was analyzed because we predicted that vowel duration would be more susceptible than consonant duration to timing deficits in speech production due to a more open vocal tract and less constriction of airflow. For the sequential motion rate task (e.g., “puhtuhkuh”), we treated all three vowels the same, regardless of which consonant they followed. Similarly, syllable rate was treated the same across all syllables in the sequential motion rate task, regardless of which consonant was produced. These two variables, syllable duration and rate, were used as dependent variables to characterize articulatory coordination because measures of DDK rate and temporal variability have been shown to be sensitive to the presence of motor speech impairment in other neurodegenerative conditions such as amyotrophic lateral sclerosis (ALS; Rong et al., 2018). Although it is not known if these features are similarly sensitive to the presence of dysarthria in ataxia, a goal of this study is to see if they are related to judgments of speech naturalness in ataxia.

Statistical Analysis

Statistical analyses were conducted with R Version 4.3.2 (R Core Team, 2022) using RStudio Version 2023.09.1 (Posit Team, 2022). Statistical code can be accessed as an RMarkdown file here, https://osf.io/wz8gk/. Eight Bayesian mixed effects models were run using Stan modeling language (Carpenter et al., 2017) and the R package “brms” (Bürkner, 2017). Bayesian modeling was chosen in contrast to frequentist modeling because of the flexible ability to define hierarchical models that include maximal random effect structure as recommended by Barr et al. (2013). For all eight models, weakly informative priors were specified for all model parameters. All models included maximal random effect structures except for the DDK models (described below), including random intercepts and slopes by speaker, allowing the fixed effects to vary by speaker.

The first two models analyzed the pitch task. The goal of the first model was to determine whether there was a group difference in the coefficient of variance in F0 (CV_F0) between the ataxia and control groups. The goal of the second model was to examine the relationship between CV_F0 and naturalness ratings only for the speakers with ataxia. Both models used regularizing Gaussian priors (μ = 0, σ = 5), signifying that we assumed no effect of group or naturalness ratings on CV_F0. For the random effects, a half Cauchy distribution was used for the standard deviation (μ = 0, σ = 0.1) and an LKJ(2) distribution for the correlation. For the residual standard deviation, a half Cauchy distribution was used (μ = 0, σ = 1).

The second two models analyzed the loudness task. The goal of the first model was to determine whether there was a group difference in the coefficient of variance in intensity (CV_int) between the ataxia and control groups. The goal of the second model was to examine the relationship between CV_int and naturalness ratings only for the speakers with ataxia. Both models used regularizing Gaussian priors (μ = 0, σ = 5), signifying that we assumed no effect of group or naturalness ratings on CV_int. For the random effects, a half Cauchy distribution was used for the standard deviation (μ = 0, σ = 0.1) and an LKJ(2) distribution for the correlation. For the residual standard deviation, a half Cauchy distribution was used (μ = 0, σ = 1).

The last four models analyzed the DDK task. The goal of the first two models was to determine if there was a group difference and an effect of the DDK task (i.e., “puhpuhpuh” vs. “tuhtuhtuh” vs. “kuhkuhkuh” vs. “puhtuhkuh”) in syllable duration and DDK rate. Attempts to include maximal random effect structures failed to converge for the first two DDK models with the issue being the interaction between group and DDK task in the random slopes. We removed the interaction and instead modeled the random slopes as group plus DDK task, which allowed the model to converge for both the model on DDK rate and DDK syllable duration. The goal of the second two models was to assess the relationship between syllable duration and DDK rate for speech naturalness and DDK task only in the ataxia group. For the second two DDK models on rate and syllable duration for the ataxia group alone, the maximal random effect structure similarly failed to converge. We removed the interaction of baturalness by task for the random slopes and instead modeled task only for random slopes, which allowed the models to converge. All four models used regularizing Gaussian priors (μ = 0, σ = 10), signifying that we assumed no effect of group or naturalness ratings on DDK syllable duration or rate. For the random effects, a half Cauchy distribution was used for the standard deviation (μ = 0, σ = 0.1) and an LKJ(2) distribution for the correlation. For the residual standard deviation, a half Cauchy distribution was used (μ = 0, σ = 1).

Four sampling chains with 2,000 iterations were run for each model, with a warm-up period of 1,000 iterations. We report the 95% credible interval and probability of direction (pd) for each effect. The probability of direction is the probability that a parameter is positive or negative (Makowski et al., 2019). Given that a value of zero indicates no effect, a higher pd value indicates a greater probability that the effect is greater than zero. The 95% credible interval means that we are 95% certain that the true value lies within the specified interval. We determine whether there is compelling evidence for an effect by whether the 95% interval overlaps with zero and whether pd is greater than 95%.

Results

Pitch Task

In Figure 1, box plots display the difference in pitch variability for the speakers with ataxia versus control participants in which the speakers with ataxia had greater pitch variability than the control participants. Contingent on the data and model, there is compelling evidence that pitch variability was higher for the speakers with ataxia (β = 0.24, 95% credible interval [0.08, 0.40], pd = 99.75%). Overall, speakers with ataxia had greater pitch variability in the pitch task than the control speakers.

Figure 1.

Pitch variability by group as measured by the coefficient of variation in F0 (Pitch CV) in cents. The box and whiskers plot shows the median pitch CV values (middle dark line), the interquartile range (outline of the box), and values outside of the interquartile range (dots).

Figure 2 displays the negative relationship between pitch variability in the pitch task and naturalness ratings for speakers with ataxia. Contingent on the data and model, there is compelling evidence for this effect in which speakers with ataxia who were rated as having more natural-sounding speech had less pitch variability in the pitch task (β = −0.13, 95% credible interval [−0.25, −0.01], pd = 98.15%, Bayesian R² = .34). Overall, speakers with ataxia with greater pitch variability were rated as having less-natural speech.

Figure 2.

Pitch variability (as measured by the coefficient of variation in F0 [Pitch CV] in cents) compared with perceptual naturalness ratings on an equal-appearing interval scale of 1–7 for the speakers with ataxia. Naturalness ratings of 1 indicated severely reduced naturalness, and ratings of 7 indicated normal perception of naturalness. The blue line indicates the trend line for the linear model between pitch CV and naturalness ratings with a 95% confidence interval (gray shading).

Loudness Task

In Figure 3, box plots display the mean intensity (dB) for each step in intensity in the FDA-2 loudness task. As can be seen from this figure, both groups increased the intensity of each step in similar patterns. Figure 4 shows box plots of CV_int, demonstrating no robust group difference for loudness variability. Contingent on the data and model, there is no compelling evidence for a group difference in loudness variability (β = 0.04, 95% credible interval [−0.10, 0.17], pd = 70.93%). Overall, both groups were comparable in their loudness variability.

Figure 3.

Box plot of median intensity (dB) for each increasing step in loudness from the Frenchay Dysarthria Assessment–Second Edition (FDA-2) loudness task by group. According to task instructions from the FDA-2, each step from the counting task of 1–5 (x axis) should increase in intensity (y axis). The box and whiskers shows the median line for intensity (dB; dark blue and yellow lines) and interquartile ranges (outline of the box), both the ataxia group (blue) and the control group (yellow). AD = Ataxic dysarthria group.

Figure 4.

Loudness variability by group as measured by the coefficient of variation (CV) in intensity (i.e., loudness CV) in dB. The box and whiskers plot shows the median loudness CV values (middle dark line) and the interquartile range (outline of the box).

Figure 5 displays the flat relationship between loudness variability in the loudness task and naturalness ratings for speakers with ataxia. Contingent on the data and model, there is no compelling evidence for a relationship between loudness variability and naturalness ratings in ataxia (β = −0.01, 95% credible interval [−0.11, 0.10], pd = 54.73%, Bayesian R² = .14). Overall, there was not a relationship between loudness variability and speech naturalness in ataxia.

Figure 5.

Loudness variability (coefficient of variation [CV] in intensity [i.e., loudness CV] in dB, y axis) by naturalness rating (x axis) for the speakers with ataxia. Naturalness ratings of 1 indicated severely reduced naturalness, and ratings of 7 indicated normal perception of naturalness. The blue line indicates the trend line for the linear model between loudness CV and naturalness ratings with a 95% confidence interval (gray shading).

DDK Task

In Figure 6, box plots display the rate (syllables/second) for each DDK task by group. As can be seen from this figure, the control group produced each DDK task at a faster rate than the ataxia group. Contingent on the data and model, there is compelling evidence for a group difference in DDK rate (β = 2.48, 95% credible interval [2.07, 2.92], pd = 100%). On average, syllable rate was 3.20 syllables/second (SE = 0.15) for ataxia participants and 5.61 syllables/second (SE = 0.13) for the control participants. There was also an overall robust effect of DDK task on syllable rate (credible interval < > 0 and pd > 99%). For both groups, the “kuh” task had the slowest rate at 3.91 syllables/second (SE = 0.09), followed by “tuh” at 4.34 syllables/second (SE = 0.09), “puh” at 4.54 syllables/second (SE = 0.09), and “puhtuhkuh” at 4.83 syllables/second (SE = 0.09). All pairwise contrasts were statistically robust (credible interval < > 0 and pd > 99%).

Figure 6.

Diadochokinetic (DDK) rate (syllables/second, y axis) by group (x axis) and DDK task (faceted columns). The box and whiskers plot shows the median DDK rate (middle dark line), the interquartile range (outline of the box), and the values outside of the interquartile range (dots) for the ataxia group (blue) compared with the control group (yellow).

Figure 7 displays the relationship between speaker naturalness rating for the ataxia participants and syllable rate in the DDK task. Overall, there was a robust relationship between syllable rate and speech naturalness (β = 0.36, 95% credible interval [0.10, 0.62], pd = 99.65%, Bayesian R² = .975). Participants who produced slower syllable rates in the DDK task were rated as less natural during conversational and passage reading tasks. There was also a robust interaction between naturalness ratings and DDK task (credible interval < > 0 and pd > 99%). Syllable rate for “puhtuhkuh” had the strongest relationship with speech naturalness ratings, followed by “puh,” “tuh,” and finally “kuh.”

Figure 7.

Diadochokinetic (DDK) rate (syllables/second, y axis) by naturalness rating (x axis, 1 = highly impaired, 7 = normal) for the speakers with ataxia by DDK task (“puh” is blue, “tuh” is yellow, “kuh” is gray, and “puhtuhkuh” is red). The solid line per color indicates the trend line for the linear model between DDK rate and naturalness ratings with a 95% confidence interval (color shading).

Figure 8 displays the effects of the participant group and DDK task on syllable duration (seconds). Contingent on the data and model, there is compelling evidence for a group difference in DDK syllable duration (β = −0.08, 95% credible interval [−0.10, −0.06], pd = 100%). On average, syllable duration was longer for ataxia participants (M = 166 ms, SE = 80) than for the control participants (M = 80 ms, SD = 40). There was also an overall robust effect of DDK tasks on syllable duration (credible interval < > 0 and pd > 99%). The “kuh” task had the longest syllable duration of 142 ms (SE = 6), followed by “tuh” at 130 ms (SE = 6), “puh” at 117 ms (SE = 6), and puhtuhkuh at 103 ms (SE = 6). All pairwise contrasts were statistically robust (credible interval <> 0 and pd > 99%).

Figure 8.

Diadochokinetic (DDK) syllable duration (seconds) by group (x axis) and DDK task (faceted columns). The box and whiskers plot shows the median DDK syllable duration (middle dark line), the interquartile range (outline of the box), and the values outside of the interquartile range (dots) for the ataxia group (blue) compared with the control group (yellow).

There was also a robust interaction between the participant group and DDK tasks (credible interval < > 0 and pd > 99%). Essentially, ataxia participants had longer syllable duration across the four DDK tasks compared to control participants. Within the ataxia group, “kuh” was produced with the longest duration (M = 197 ms, SE = 8), followed by “tuh” (M = 175 ms, SE = 8), then “puh” (M = 157 ms, SE = 8), and “puhtuhkuh” (M = 137 ms, SE = 8). All interactions between tasks within the ataxia group were robust (credible interval < > 0 and pd > 99%). Within the control group, all interactions were robust except for the difference between “tuh” (M = 84 ms, SE = 8) and “kuh” (M = 88 ms, SE = 8, credible interval [−0.02, 0.01], pd = 75.98%). Overall, ataxia participants had longer syllable duration across DDK tasks than control participants. Within the ataxia group, “kuh” was produced with the longest duration, followed by “tuh,” then “puh,” and finally “puhtuhkuh.” Within the control group, there were less robust effects by task.

Figure 9 displays the relationship between speaker naturalness rating for the ataxia participants and syllable duration in the DDK task. Overall, there was a robust relationship between syllable duration and speech naturalness (β = −0.04, 95% credible interval [−0.06, −0.01], pd = 99.15%, Bayesian R² = .44). Participants with ataxia who produced longer syllables in the DDK task were perceived as less natural during conversational and passage reading tasks. There was no robust interaction between the naturalness rating and DDK task, indicating that syllable duration across all tasks was related to speech naturalness.

Figure 9.

Diadochokinetic (DDK) syllable duration (seconds) by naturalness rating (x axis, 1 = highly impaired, 7 = normal) for the speakers with ataxia by DDK task (“puh” is blue, “tuh” is yellow, “kuh” is gray, and “puhtuhkuh” is red). The solid line per color indicates the trend line for the linear model between DDK syllable duration and naturalness ratings with a 95% confidence interval (color shading).

Discussion

The goal of this study was to evaluate the relationship between the perceptual rating of speech naturalness and objective measures of pitch variability, loudness variability, and rate control to determine whether such objective measures may serve as potential tools for assessment of ataxic dysarthria and for validation of speech naturalness judgments. We hypothesized that speech coordination impairments in cerebellar ataxia are reflected in simple, commonly used speech assessment tasks, such as pitch glides, loudness step tasks, and DDK rates. We predicted that there would be greater variability in pitch, loudness, and rate control in the ataxia group compared to neurologically healthy control participants. Furthermore, we predicted that greater variability in all three of these acoustic measures would be robustly related to the perception of speech naturalness. The findings of this study generally supported our hypothesis and have implications for a more objective assessment of speech naturalness in clinical practice. Pitch variability (as indexed from the variability in pitch steps in the FDA-2 pitch glide task), DDK rate, and DDK syllable duration were all robustly correlated with speech naturalness ratings. Speakers with ataxia with lower naturalness ratings had greater pitch variability, longer DDK vowel segment durations, and a slower DDK rate in comparison to speakers with higher naturalness ratings. These results demonstrate that objective measures of pitch variability, DDK segment duration, and DDK rate are reflective of perceptual measures of speech naturalness.

A significant finding from this study is that increased variability in pitch control correlated with reduced speech naturalness. This finding should be interpreted within the broader context of the literature on ataxic dysarthria and cerebellar control mechanisms. There is evidence that the cerebellum is involved with feedback control of pitch during online speech production, which could explain some of the prosodic impairments in ataxic dysarthria. When pitch auditory feedback is manipulated in people with ataxic dysarthria through the pitch auditory feedback perturbation paradigm, participants overcompensate for the perceived pitch errors with larger responsive changes in pitch than neurotypical control participants (Hilger, 2020; Houde et al., 2019; Li et al., 2019). In application to everyday speech, this finding implies that variability in pitch could be due to impaired auditory feedback control to adjust intonation. One of the classic features of ataxic dysarthria is ambiguous phrasal stress and boundary (Darley et al., 1969; Kent et al., 2000; Lowit et al., 2014), which are signaled through changes in pitch (in addition to loudness and timing) in English (Cole, 2015). When these prosodic features are less salient, it is likely to reduce the perception of speech naturalness, although this is an important future research question to explore. The current study provides some evidence for this relationship between pitch variability and speech naturalness. Participants who had more variability when increasing pitch during the pitch glide task were also rated as having less natural speech in connected speech samples. To consistently produce evenly spaced notes in a pitch scale, a person would need accurate auditory feedback control to identify the pitch being produced and correctly scale the increase of pitch needed for the next note in the scale. Based on the results of our study, it is likely that the increased variability in pitch during this simple pitch glide task reflects impaired pitch auditory feedback control for speech in general, reducing speech naturalness. A future study to confirm this finding would involve assessment of pitch variability in connected speech in these same participants to determine whether those who exhibit variability in the pitch glide task also exhibit greater variability in connected speech and if these measures continue to robustly relate to speech naturalness.

Another consideration is how pitch variability is operationally defined and how it is measured. In the present study, a speaker was considered to have high pitch variability if they produced inconsistent steps between notes on a seven-note scale. A speaker must rely heavily on auditory and somatosensory feedback to produce evenly spaced notes on a scale. Given that the cerebellum plays an important role in the feedback control system (Hilger, 2020; Houde et al., 2019; Li et al., 2019), this particular index of pitch variability is expected to be sensitive to the changes in motor control that occur in ataxia.

A study by Hertrich and Ackermann (1993) found that intra-utterance variation in F0 was in the normal range for speakers with Friedreich's ataxia. This finding contrasts with our study in which pitch variability in the FDA-2 task was robustly different from control participants and correlated with speech naturalness. Task choice likely explains this difference in findings, and that different tasks were selected based on the goals of each study. First, the production task in the study of Hertrich and Ackermann (1993) was an utterance production task, whereas the present study utilized the FDA-2 pitch glide task. The range of F0 differs considerably between these two tasks. To produce utterances, a person will vary their F0 around a comfortable level with slight variation for emphasis or syntax. In a pitch glide task, a person must sing notes that are likely out of a comfortable F0 level, requiring greater phonatory coordination. Therefore, the task used in our study, the FDA-2 pitch glide, was likely a more complex pitch task for the participants with ataxia, requiring greater coordination and resulting in greater variability. Another factor in the conflicting findings is the difference in the goals of each study. In Hertrich and Ackermann's (1993) study, they were interested in whether pitch control was in the normal range. In the current study, we were interested in whether pitch variability would be robustly related to speech naturalness, and we did not investigate whether these values were in a normal range for the task (which, to our understanding, no values currently exist). Therefore, a future direction is to conduct a similar analysis to that of Hertrich and Ackermann (1993) by comparing utterance-level F0 variability with perceptual speech naturalness measures to see if a similar relationship unfolds as the one found in this current study.

The one task that did not have a robust relationship with speech naturalness was loudness variability, as indexed by the loudness step task in the FDA-2. There was no group difference in loudness variability between the ataxia and control groups. Within the ataxia group, the relationship between loudness variability and speech naturalness was negligible. A likely explanation is that this task was too structured and simplistic for the participants with ataxia as compared with everyday naturalistic speech. Loudness control is easier when counting from 1 to 5 with increasing levels of intensity, as opposed to controlling loudness across natural utterances, because the loudness step task has clearer targets. At the same time, the targets for this task are not as stringent as the targets for the pitch glide task in which speakers generally attempt to reproduce precise notes on a musical scale. This difference in task structure allows for greater variability between intensity steps for both speakers with ataxia as well as controls, which might explain why this measure of loudness variability did not robustly correlate with speech naturalness. Overall, this finding demonstrates that speakers with ataxia can coordinate breath control and phonation for appropriate loudness during structured tasks and thus have the potential to improve loudness control during more unstructured tasks in speech therapy. A future direction for this work will be to assess loudness variability and speech naturalness in more complex tasks.

This null finding for loudness variability is in contrast to previous literature that has measured increased intra-utterance variability in intensity for participants with Friedreich's ataxia specifically (Hertrich & Ackermann, 1993). This difference in findings makes sense based on task complexity. In our study, the FDA-2 loudness step task was a simple, structured task; in the Hertrich and Ackermann (1993) study, participants produced more complex utterances. As stated previously, it is likely easier for speakers with ataxia to coordinate respiratory and phonatory subsystems for appropriate loudness variation during simple tasks.

Between-groups comparisons of DDK production replicated previous research (Brendel et al., 2015; dos Santos et al., 2023; Portnoy & Aronson, 1982; Wang et al., 2009). We found that participants with ataxia performed more poorly on the DDK tasks than the control participants. Specifically, participants with ataxia had longer syllable duration and slower syllable rate across the DDK tasks compared to control participants. This finding is unsurprising given the significant role of the cerebellum in movement coordination and provides continual support for the application of the DDK task in the assessment of dysarthria in ataxia.

Another expected finding from this study was that participants in both groups exhibited decreased performance for some DDK tasks compared to others. Specifically, participants used a slower rate with longer syllable durations for the “kuh” and “tuh” tasks compared to the “puh” and “puhtuhkuh” tasks. This task effect is due to articulatory coordination and control differences for the lips, tongue tip, and tongue dorsum. Speakers have greater control and coordination over lip movements than tongue movements partially because coupling of the tongue and jaw requires greater muscular coordination (DePaul & Brooks, 1993); therefore, repeating the “puh” syllable may be produced more rapidly with a shorter duration than the alveolar and velar lingual consonants “tuh” and “kuh.” The tongue tip is also better coordinated than the tongue dorsum, partly because it is less bulky and also because it has more sensory fibers for proprioceptive feedback; these features also make it easier to rapidly repeat “tuh” compared with “kuh” (Green & Wang, 2003; Rong, 2020; van Lieshout, 2017). The AMR and SMR tasks are specifically useful for identifying these deficits in motor execution, because increasing the rate of movement can weaken existing coordination patterns of articulatory movements (van Lieshout, 2017). Last, the “puhtuhkuh” sequence may have been faster than “tuh” and “kuh,” because it is easier for speakers to coordinate sequential movements rather than repeated movements. These task differences were noted for both groups of speakers in this study with one exception: Control participants did not have a significant difference in syllable duration between “tuh” and “kuh.” A potential explanation is that coordination and performance differences between these two tasks are minimal among individuals with intact tongue control capabilities.

The interesting and important finding from this study is that, for speakers with ataxia, pitch glide and DDK performance are highly correlated with perceptual speech naturalness ratings. This is important because speech naturalness is a perceptual measure that is subjective and can be highly variable from clinician to clinician based on their clinical experiences as well as the listening conditions during the evaluation. This study shows that objective measures, such as segment duration and rate in DDK, are highly correlated with perceptual measures of speech naturalness. Similarly, the objective measure of pitch variability in the FDA-2 pitch glide task was highly correlated with the perceptual measure of speech naturalness.

Clinically, SLPs can combine perceptual measures of speech naturalness and objective measures of pitch variability and DDK production to obtain a comprehensive measure of speech ability when assessing patients with ataxic dysarthria. This finding is particularly relevant given that very few studies have attempted to correlate speech naturalness with other characteristics of speech production. For example, in their systematic review of 63 studies published between 1990 and 2018 that measured speech naturalness, Klopfenstein et al. (2020) found that 70% of studies did not attempt to correlate naturalness with other variables that were measured. By directly examining such correlations, the present study informs our understanding of what specific variables may contribute to listeners' judgments of impaired speech naturalness. These findings can improve SLPs' assessments of speech naturalness and inform their predictions of how specific treatment approaches might impact naturalness, whether or not this is their primary target of intervention.

Future directions of this work include acoustic analyses of pitch, loudness, and articulatory variability during spontaneous speaking tasks to further characterize the relationship between acoustic speech characteristics and perceptual judgments of speech naturalness. Additionally, the incorporation of these objective measures into motor speech assessment and treatment could potentially serve as a tool for biofeedback and progress monitoring during intervention. Results of the present study can be applied to future research evaluating the effectiveness of speech therapy interventions for ataxia that utilize biofeedback and combined perceptual and instrumental approaches to intervention.

There are limitations within the present study that should be considered in the interpretation and implementation of findings. First, dysarthria type and severity were determined by a single rater, and there was no protocol for identifying potential mixed dysarthrias among the group of participants. Although all participants did exhibit a primary ataxic dysarthria, hereditary ataxias can lead to neuronal loss beyond the cerebellum (Taroni & DiDonato, 2004), which may lead to the development of mixed dysarthrias over time. The presence of mixed dysarthria may impact clinical profiles of impairment and performance on speech tasks. Future studies could benefit from a protocol for identifying mixed dysarthrias in conjunction with a larger sample size spanning a broad range of ataxia etiologies. Next, speech naturalness was rated using an EAI rating scale from 1 to 7. Although there is evidence from the stuttering and tracheoesophageal speech literature to suggest that speech naturalness can be considered a metathetic dimension of speech that can be evaluated reliably using interval rating scales (Eadie & Doyle, 2002; Metz et al., 1990), some evidence suggests that visual analog scales and/or direct magnitude estimation scales may be better suited for evaluation of some speech characteristics such as nasality (Whitehill, 2002; Zraick & Liss, 2000) and intelligibility (Abur et al., 2019). Further exploration into valid measures of speech naturalness is warranted, particularly within motor speech research (Klopfenstein et al., 2020). Finally, this study evaluated acoustic characteristics of pitch, loudness, and rate control in structured “quasi-speech” tasks, which were then correlated with naturalness ratings derived from connected speech samples. The use of “quasi-speech” or “speechlike” tasks for the purpose of motor speech evaluation has undergone scrutiny (e.g., Weismer, 2006; Ziegler, 2003). A future direction of this work is to relate acoustic analyses of pitch, loudness, and articulatory control in connected speech tasks to speech naturalness ratings to better understand the complexities of speech motor control outside of maximum performance or quasi-speech tasks.

Conclusions

The goal of this study was to determine whether objective measures from the pitch and loudness tasks in the FDA-2 and acoustic analyses of DDK production are related to perceptual measures of speech naturalness in cerebellar ataxia. We found a strong relationship between speech naturalness ratings and objective measures including pitch variability in the FDA-2 pitch glide task, DDK syllable duration, and DDK syllable rate, supporting our hypothesis. Interestingly, there was a null relationship between loudness variability in the FDA-2 loudness step task and speech naturalness, indicating that this measure is not a good objective measure of this perceptual rating. For the DDK task specifically, participants with ataxia produced syllables at a slower rate and with longer syllable durations than control participants, and all participants demonstrated better motor control during lip-movement tasks (e.g., “puh”) and sequential repetition tasks (e.g., “puhtuhkuh”) compared to repeated tongue movement tasks (e.g., “tuh” and “kuh”). Overall, SLPs can incorporate both perceptual measures of speech naturalness and acoustic measures of pitch variability and DDK performance for a comprehensive evaluation of ataxic dysarthria.

Data Availability Statement

Statistical code can be accessed as an RMarkdown file here, https://osf.io/wz8gk/. Data are available upon request.

Funding Statement

This research was funded by National Institute on Deafness and Other Communication Disorders Grant F31 DC017877-01A1 and the Council of Academic Programs in Communication Sciences and Disorders.