Sports Related Concussion Impacts Speech Rate and Muscle Physiology

Abstract

Objective:

Establish objective and subjective speech rate and muscle function differences between athletes with and without sports related concussion (SRC) and provide a potential means of evaluating motor speech in SRC populations.

Methods:

Over 1,110 speech samples were obtained from 30, 19-22 year-old athletes who had sustained an SRC within the past 2 years and 30 pair-wise matched control athletes with no history of SRC. Speech rate and muscle function were evaluated during diadochokinetic (DDK) tasks. Speech rate was measured via average time per syllable, average unvoiced time per syllable, and expert perceptual judgement. Speech muscle function was measured via surface electromyography over the obicularis oris, masseter, and segmental triangle. Group differences were assessed using MANOVA, bootstrapping and predictive ROC analyses.

Results:

Athletes with SRC had slower speech rates during DDK tasks than controls as evidenced by longer average time per syllable (F(1, 52) = 11.072, p =.002, [95% CI : .01 to .04]), longer average unvoiced time per syllable (F(1, 52) = 16.031, p < .000, [95% CI : .01 to .029] and expert judgement of slowed rate (F(1, 22) = 9.782, p = .005, [95% CI : .163 to .807]). Rate measures were predictive of concussion history. Further, athletes with SRC required more speech muscle activation than controls to complete DDK tasks (F(1, 3) = 17.12, p =.000, [95% CI: .003 to .006]).

Conclusion:

Clear evidence of slowed speech and increased muscle activation during the completion of DDK tasks in athletes with SRC histories relative to controls. Future work should examine speech rate in acute concussion.

INTRODUCTION

Speech production is a complex neuromotor process dependent on millisecond timing and coordination of a large number of cortical and subcortical structures and muscle groups^1–4. The temporal features of speech motor control are particularly sensitive to subtle changes in neural function and have been leveraged as early indicators of neurological impairment associated with Parkinson’s disease⁵, Huntington’s disease^6,7, Amyotrophic Lateral Sclerosis^8,9, Chronic Traumatic Encephalopathy (CTE)^10,11 and traumatic brain injury (TBI)^12,13. Speech timing changes may also be an important early indicator of sports-related concussions (SRC), a subset of TBI.

To the untrained listener, changes in speech timing and coordination are often perceived as “slurred speech.” Slurred speech is widely recognized by contact sport referees as an on-field sign that an athlete has sustained a SRC¹⁴. The most frequently cited sideline assessments and guidelines list slurred speech as a physical sign of acute SRC and encourage that speech be monitored at follow-up^15–18. Surprisingly, no operational definition of slurred speech as it pertains to SRC has been established and no protocol exists to assess speech motor control in SRC. Establishing a meaningful and efficient speech evaluation to inform current comprehensive SRC assessments could improve diagnostics and prognostics for affected athletes, as underscored by the call for speech analyses in SRC by the American Medical Society for Sports Medicine’s (AMSSM)¹⁷.

Diadochokinetic (DDK) speech tasks allow for the easy and efficient evaluation of speech timing attributes^19–22. Rate of production and pause time for DDK speech tasks can be measured objectively and subjectively using short recordings to inform measurements of an athlete’s speech characteristics post SRC. Speech rate during DDK tasks has shown promise for distinguishing SRC from controls²¹.

Preliminary findings from a small mixed sample of patients with SRC and non-SRC mild TBI found slower DDK rates in patients compared to controls¹³. The present study builds on this preliminary evidence to comprehensively quantify DDK speech rate across objective and subjective measures and to examine, for the first time, any articulatory muscle differences using surface electromyography (sEMG), in former athletes with SRC histories compared to controls. We aimed to determine a clinically relevant protocol for the evaluation of speech production and speech muscle physiology in the context of SRC. Specifically, the objective of the current study is to provide a potential means of evaluating speech motor ability in previously concussed populations and establish persistent objective and subjective group differences in speech production between athletes with a history of concussion and those without.

METHODS

Enrollment

Under protocol approval from the Michigan State University Human Research Protection Program, all 60 participants were recruited consented from a student research participation pool from the University’s College of Communication Arts and Sciences. Potential participants were selected for further screening based on survey responses to determine their eligibility for inclusion in this study. The 30 participants with SRC histories (aged 19-22 years) were current or former athletes at the senior high school or collegiate level, with English as their first language, and a negative self-reported history of dyslexia and neurological disorders. Additionally, all participants indicated during screening that acute symptomology had resolved, with no lingering sequalae and specifically, no appreciable speech or voice deficits.

The participants with SRC sustained their most recent concussion within the last 2 years, as motor function may be affected long-term in this population²³. Participants must have been diagnosed by an MD or athletic trainer, had not been hospitalized for their injury, and had not lost consciousness for more than 20 minutes at the time of injury. One SRC participant’s data was excluded from analysis after the participant later reported the presence of an exclusionary co-morbid neurological disorder. The 30 control (CON) participants were recruited as pairwise matches for the SRC group for height (within 10in.), weight (within 20lbs.), age (within 3 years), education (within 1 grade level) and gender. CON were also made up of both current and former athletes.

Procedures and Analyses

Speech Tasks

Participants performed a series of speech and oromotor tasks after being fitted with a microphone and sEMG sensors over the obicularis oris, masseter, and segmental triangle (see Recording Equipment below). The speech DDK tasks were three Alternating Motion Rate (AMR) tasks on the individual syllables puh, tuh, and kuh; and three Sequential Motion Rate (SMR) tasks on puh-tuh-kuh; and the words pattycake and buttercup (see Table 1). Tasks were demonstrated for clarity. Participants were instructed to produce DDKs at their maximum rate. Participants practiced the DDK tasks before recording.

Table 1.

List of speech tasks and the timing, perceptual and sEMG analyses.

Tasks	Objective Analysis	Subjective Analysis
DDK REPETITIONS: AMR syllables – Single syllable puh, tuh, kuh syllables (repeated 10 times each) SMR syllables non-word - puh-tuh-kuh (repeated 10 times) SMR real words - pattycake & buttercup (repeated 10 times each)	ACOUSTIC TIMING: Average time per syllable - TPS: [total time to produce 10 repetitions in seconds / 10 repetitions] Average unvoiced time per syllable - UTPS: [(total time to produce 10 repetitions in seconds – total voicing time in 10 repetitions) / 10 repetitions] sEMG: Average sEMG (RMS) - over 10 repetitions	Articulatory Precision – “Are they producing the target syllable?” Articulatory Rate – “Compared to a ‘normal’ modulus, is speech faster or slower?” Rhythmic Consistency – “Is the length of each syllable and the time between syllables uniform?”

Recording Equipment

The speech audio signals were recorded using an omnidirectional head-mounted microphone (M80, Glottal Enterprise, Syracuse, NY, USA) connected to a digital audio recorder (Roland R-05, 44,100 Hz, 24-bit, wav file). The microphone was placed 3–4 cm from the side of the mouth (outside the primary airstream). A wireless micro sEMG system (Trigno; www.delsys.com/products/wireless-emg) was used to detect speech motor muscle activation and electrical activity. A PowerLab 8/35 (ADInstruments, New South Wales, Australia) interfaced with the sEMG system and the digital audio recorder to simultaneously sample all signals which were recorded on a laptop computer running LabChart 10.1 software. The sEMG sensors (see Figure 1) were attached to the skin on the obicularis oris (OO), masseter (MAS), and submental triangle (ST; where tongue muscle activation can be measured) with manufacturer-supplied hypoallergenic adhesives. Sensor placement was guided by standard research practices for recording both normal and disordered speech^24–26. Prior to sEMG placement, the location was prepared per standard practices (e.g. cleaned, lightly abraded)^24,26.

Figure 1.

Sample acoustic waveform and timing results obtained from control male (A.) and SRC history female (B. & C.) using automated timing and sEMG analysis. The green wave forms in figures A and B are the unfiltered acoustic signal while the blue waveforms are the filtered signal. The space between green lines on the blue waveforms indicates the estimated voiced time. Figure 1C. also depicts the placement of sEMG units over the obicularis oris, massater, and submental triangle and the position of the head-mounted microphone.

Objective Speech Measures: Timing

We created three schedules for task completion, each with its own randomized order, and then randomly assigned a schedule to each participant with SRC. The control participants were given the same schedule as the participant with SRC with whom they were matched. All participants completed three practice attempts of each task to establish familiarity with the protocol and ensure correct performance. Segmentation was performed in Audacity audio manipulation software (https://www.audacityteam.org/) and analyzed using a fundamental filter between 70 Hz and 450 Hz for females, and between 50 Hz and 350 Hz for males. Segmentation and filtering was carried out in a custom, automated Matlab script following general guidelines described by Salvatore et al and other recent publications^13,27,28. Specifically, our scripts pay particular attention to the start and end of acoustic signals, ensuring that pauses not associated with the speech signal were eliminated, thus increasing the fidelity of our processed signals. We further applied a loudness filter between 5-75 decibels (dB), the loudness range where speech is most likely to be present (Figure 1). Thus, signal above or below these filter ranges were considered pause or unvoiced speech. To ensure that speech was properly segmented, the results of the automated analyses were all manually reviewed for accuracy by the first author of this experiment. Three sets of ten correct repetitions were counted during each AMR and SMR task using audio recordings and, when necessary, spectral and waveform settings.

Estimated Average Time Per Syllable (TPS):

Following previously established protocols which eliminate the first DDK productions, providing a more accurate depiction of participant abilities, final segmentations included the 2^nd-11^th correctly produced target¹³. These ten productions were used in the statistical analyses described below. Timing measures for each set of ten repetitions were then averaged and used in statistical analysis. These segmented recordings became the basis for all timing, acoustic and perceptual analyses. To calculate the average time per production, the total length of each file containing the ten repetitions of each DDK task was divided by the total number of syllables. Only correctly articulated multisyllabic DDK productions were included in our time by count analysis.

Average Unvoiced Time Per Syllable (UTPS):

The length of unvoiced time was calculated as the time in which voicing (measurable vibration of the vocal folds in Hertz (Hz)) was not detected during DDK productions divided by the number of syllables (10 for the AMR’s and 30 for the SMR’s). Speech timing analysis was automated using a Matlab script described by Bottalico and colleagues²⁷. Briefly, we used the aforementioned dB and frequency calculation

Objective Speech Measures: Surface Electromyography Measures

Root Mean Squared (RMS-sEMG) –

RMS-sEMG was used to quantify the amount of electrical activity in the motor units of the muscles under study. There is a positive linear relationship between EMG output and imposed load of facial muscles^29,30. This measure was applied to muscles in the jaw, lip and base of tongue in the current study to determine potential differences in electrical activity of these muscles between groups during the performance of DDK tasks. Direct current signal offset was filtered so that signal means were zero. A bandpass filter was applied between 20 and 450Hz and windowed at 500ms in order to remove movement and equipment artifact and noise (Figure 1)²⁶.

Perceptual Ratings

Measures to be judged were explained to all expert speech-language pathologist (SLP) raters via phone (2) or video (6) conferencing. Expert raters were defined as clinically certified speech language pathologists, with professional experience in treating voice and speech motor disorders, who were trained by the first author of this manuscript. Following live instruction, SLP raters were given a written copy of instructions, examples and contact information in the event of questions during evaluation. Raters were blinded as to the condition (SRC or CON) of the voices being rated. Each rater evaluated between 150 and 200 randomly selected speech samples with at least 20% of the recordings randomly repeated with and between raters in order to analyze reliability.

Articulatory Precision and Rhythmic Consistency –

Defined measures of articulatory precision and rhythmic consistency were used to rate the accuracy and precision of concussed speakers³¹. Samlan and Weismer defined articulatory precision as “the accuracy with which the speaker produced the consonant and vowel targets.” They then defined rhythmic consistency as “the uniformity of the length of each segment and the time between segments.” SLP raters were told to use these definitions in order to make their judgements. To further assist in the completion of judgements, raters were provided a “normal” modulus to compare to each recording in terms of articulatory precision and rhythmic consistency.

Rate –

Previous research found changes in basic parameters of the voices of individuals following a concussion, primarily in their fundamental frequencies and variability^32,33. However, no study to date has described a formal subjective evaluation of rate in this population. It was anticipated that trained SLP listeners would note increased slowness demonstrated by a reduced rate in those with SRC histories compared to speech of those without SRC histories. SLP judges were provided a “normal” modulus to compare to each recording in terms of rate (see Table 1).

Statistical Analyses

Three separate multiple analyses of variance (MANOVA) models were used to compare objective speech rate, perceptual analyses, and sEMG output of articulatory muscles between those with concussion histories and those without. Due to the independent nature of the data contained in our 3 MANOVA models, it was determined unnecessary to adjust p values for multiple comparisons. To determine the potential predictive nature of our measures, AUROC was applied to results.

RESULTS

Participant demographics are detailed in Table 2. Successful control and matching were demonstrated in the lack of significant differences between groups in height, weight, year of schooling, and age. Twenty-six males and 34 females participated, though one male was eliminated after recording due to a screening violation. While previous literature in SRC has demonstrated sex differences, initial statistical models did not demonstrate any effect and thus, were not included in our models.

Table 2.

Participant demographic and pairwise matching results indicating no significant differences between groups.

Factor	No History	Concussion History	P value (±SD)
Height (in.)	68.05	67.3	.12 (± 4.06)
Weight (lbs.)	155.8	153.9	.63 (± 32.39)
Age (YOB)	1997.5	1997.7	.25 (± 1.06)
Self-Identified Race/Ethnicity	3% Asian; 7% Black/African American	6% Asian; 10% Black/African American	NA
Average Time Since Injury	NA	11 months (± 11.4)	NA
Average Year of Education	1st year undergraduate (13y)	1st year undergraduate (13y)	NA
Sex	17F; 13M	17F; 13M	NA

Objective Speech Rate Analysis

A MANOVA using average times per syllable (TPS) and unvoiced times per syllable (UTPS) of all DDK tasks as independent variables was completed. Mauchly’s Test of Sphericity revealed the sphericity assumption was violated for each of the timing variables used in this model (p<.001 for each variable), therefore a Greenhouse-Geisser correction was applied. The model showed a between-subjects main effect of SRC on measures of TPS (F(1, 52) = 11.072, p =.002, np2 =.18 [95% CI : .01 to .04]) and UTPS (F(1, 52) = 16.031, p < .000, np2 = .24 [95% CI : .01 to .029]. To determine which specific DDK tasks were significantly different between groups, bootstrapping (5,000 reshuffles) of control and SRC data was performed and results are reported in Table 3. Results showed slower average TPS for those with a history of SRC compared to controls during all DDK productions. Additionally, longer average UTPS was noted during all DDK tasks except kuh (p= 0.051) and pattycake (p= 0.135) productions (Table 3). Figure 2 illustrates mean differences, effect sizes and confidence intervals in average TPS and UTPS using Gardner-Altman plots³⁴. Figures 2A and 2B illustrate differences in times per syllable and unvoiced times per syllable, respectively.

Table 3.

Bootstrapping results with Welsch’s t-statistic provided for timing analyses and sEMG data obtained during DDK tasks demonstrated clear timing differences between SRC and control groups. Means are reported in syllables per second. Negative mean differences indicate that syllable times for those with a history of concussion were longer than those with no concussion history.

	Analysis	DDK Tasks	Welch’s t-Statistic	Mean Control	Mean SRC	Mean Dif.	Welch’s P-value	BH P-value	BH Critical Value	Cohen’s d	Low 95% CI	Hi 95% CI
Timing Analyses	TPS	Puh	−3.32	0.15	0.17	−0.02	0.001	0.002	0.022	0.5	0.21	0.74
		Tuh	−5.25	0.14	0.17	−0.04	0	0	0.003	0.88	0.56	1.13
		Kuh	−3.1	0.16	0.18	−0.02	0.002	0.003	0.028	0.47	0.17	0.72
		PuhTuhKuh	−4.02	0.16	0.19	−0.03	0	0	0.006	0.61	0.3	0.87
		Buttercup	−4.51	0.15	0.17	−0.02	0	0	0.008	0.68	0.39	0.97
		Pattycake	−4.18	0.16	0.18	−0.02	0	0	0.011	0.63	0.26	1
	UTPS	Puh	−4.39	0.09	0.1	−0.02	0	0	0.014	0.67	0.38	0.93
		Tuh	−5.88	0.08	0.11	−0.03	0	0	0.017	0.98	0.72	1.25
		Kuh	−1.95	0.1	0.11	−0.01	0.051	0.054	0.047	0.3	−0.13	0.65
		PuhTuhKuh	−4.69	0.1	0.13	−0.03	0	0	0	0.71	0.38	0.98
		Buttercup	−3.42	0.09	0.1	−0.01	0.001	0.002	0.025	0.52	0.2	0.83
		Pattycake	−1.5	0.1	0.11	−0.01	0.135	0.135	0.05	0.23	−0.1	0.53
	Perceived Rate	Puh	3.05	0.54	−0.11	0.65	0.002	0.003	0.031	−0.78	−1.25	−0.31
		Tuh	2.8	0.5	−0.11	0.61	0.01	0.014	0.036	−0.7	−1.13	−0.18
		Kuh	2.83	0.18	−0.43	0.61	0.008	0.012	0.033	−0.72	−1.21	−0.19
		PuhTuhKuh	2.38	0.14	−0.46	0.6	0.017	0.022	0.039	−0.62	−1.15	−0.07
		Buttercup	2.27	0.01	−0.42	0.43	0.021	0.025	0.042	−0.6	−1.16	−0.04
		Pattycake	2.05	0.04	−0.38	0.42	0.042	0.047	0.044	−0.52	−1.03	−0.04
sEMG	RMS Lip	Puh	−3.94	0.07	0.15	−0.08	0	0	0.004	0.59	0.41	0.41
		Tuh	−3.49	0.02	0.01	0.01	0	0	0.008	0.52	0.35	0.35
		Kuh	−2.6	0.03	0.05	−0.02	0.005	0.0075	0.033	0.39	0.19	0.19
		PuhTuhKuh	−2.85	0.06	0.12	−0.05	0.002	0.004	0.025	0.43	0.23	0.23
		Buttercup	−2.51	0.07	0.13	−0.06	0.006	0.008	0.038	0.38	0.18	0.18
		Pattycake	−4.04	0.06	0.13	−0.07	0	0	0.013	0.59	0.41	0.41
	RMS Jaw	Puh	−2.86	0.03	0.04	−0.01	0	0	0.017	0.43	0.26	0.59
		Tuh	−1.86	0.03	0.04	−0.01	0.037	0.04036	0.046	0.26	0.07	0.41
		Kuh	3.19	0.03	0.03	0	0.001	0.0024	0.021	−0.49	−0.73	−0.16
		PuhTuhKuh	−1.75	0.03	0.08	−0.04	0.003	0.00514	0	0.31	0.18	0.49
		Buttercup	−2.24	0.03	0.04	−0.01	0.007	0.0084	0.042	0.36	0.15	0.54
		Pattycake	−0.09	0.03	0.03	0	0.944	0.944	0.05	0.01	−0.28	0.31

Bootstrapping results with Welsch’s t-statistic provided for timing analyses and sEMG data obtained during DDK tasks demonstrated clear timing differences between SRC and control groups. Means are reported in syllables per second. Negative mean differences indicate that syllable times for those with a history of concussion were longer than those with no concussion history.

TPS = time per syllable; UTPS = unvoiced time per syllable.

A Benjamini-Hochberg (BH) correction was applied where corrected P-values below the BH critical values .047 represents significant results.

Figure 2.

The average TPS (A), UTPS (B), and perceived rate (C) for all DDK comparisons are shown in the above Gardner-Altman estimation plots. Control data is plotted in blue and SRC history plotted in brown; each Cohen’s d is plotted on a floating axis on the right as a bootstrap sampling distribution. Cohen’s d is depicted as a black dot; bootstrap 95% confidence intervals are indicated by the ends of the black vertical error bars.

Perceptual Analysis

A MANOVA was performed using raters’ judgments of articulatory precision, articulatory rate, and rhythmic consistency for all six DDK tasks as independent variables. No violations to the assumptions were noted in this model. A between-subjects main effect of SRC history on trained SLP judgements of speech rate (F(1, 22) = 9.782, p = .005, np2 =.308 [95% CI : .163 to .807]). The model did not detect a between-subjects main effect of SRC history on raters judgments of articulatory precision (F(1, 22) = .326, p =.574, np2 =.015 [95% CI : −.603 to .343]) or rhythmic consistency (F(1, 22) = .141, p = .710, np2 = .006 [95% CI :CI : −.32 to .416]). The percentage of interrater agreement were: 54% (fair) for articulatory precision, 74.5% (good to excellent) for articulatory rate, and 67.9% (fair to good) for rhythmic consistency³⁵. To determine which specific DDK tasks were perceived as significantly different between groups, bootstrapping (5,000 reshuffles) of control and SRC data was performed and results indicated that all DDK tasks were perceived to be significantly slower than controls (Table 3. & Figure 2C.).

SEMG ANALYSIS

Root Mean Squared (RMS)

A MANOVA was performed analyzing the average measured output of EMG units attached to the face during the production of all DDK tasks. The model showed a between-subjects main effect of SRC history on the average measures EMG output across muscles of the face (F(1, 3) = 17.12, p =.000, np2 =.118 [95% CI: .003 to .006]. To further investigate which DDK tasks produced significant differences, bootstrapping analyses of lip, jaw, and tongue muscle activation were performed comparing SRC and controls.

Lip –

The model showed a between-subjects main effect of SRC history on the average measures RMS-EMG output of the obicularis oris. Individuals with a history of SRC demonstrated significantly greater lip muscle activation compared to participants with no SRC history during all DDK tasks with the exception of tuh productions, where controls had significantly more lip muscle activation (Table 3 & Figure 3A).

Figure 3.

The average sEMG-RMS of Lip (A) and Jaw (B) muscle activation during all DDK comparisons are shown in the below Gardner-Altman estimation plots. Control data is plotted in blue and SRC history plotted in brown; each Cohen’s d is plotted on a floating axis on the right as a bootstrap sampling distribution. Cohen’s d is depicted as a black dot; bootstrapped 95% confidence intervals are indicated by the ends of the black vertical error bars.

Jaw –

The model showed a between-group difference on the average measures of RMS-EMG where those with a history of SRC demonstrated significantly greater muscle activation in the masseter muscles of participants with a SRC history compared to controls during all DDK productions with the exception of pattycake (p=.0944) productions only (Table 3. & Figure 3B).

Tongue –

The model was unable to detect differences between groups on the average measures of RMS-EMG where those with a history of concussion demonstrated greater muscle activation in the base of tongue compared to participants with no SRC history.

AUROC Analysis

An area under the receiver operating curve (AUROC) analysis was carried out to understand the predictive diagnostic ability of each of the above measures (see Figure 4). The predictive value of our TPS measure to identify patients with concussion indicates Tuh productions possess the highest discriminatory ability (moderate at Area Under the Curve (AUC)=0.77). Assuming values greater 0.136 syllable/sec indicate a positive concussion history, sensitivity (ability to identify those with a history of concussion) was high (0.86) but specificity (correctness of those classified as no concussion history) was low (0.54). For our UTPS measure, Puh had the highest discriminatory value (AUC= 0.83). With a cut-off of 0.0844 seconds/syllable, sensitivity was high (0.83) and specificity moderate (0.7). SLP’s Perception of rate during Buttercup provided the highest discriminatory value (AUC= 0.75). Applying a cut-off of approximately seven perceived units slower than the provided modulus, sensitivity was high (0.84) but specificity was low (0.62). AUROC values obtained for sEMG data were not able to reliably distinguish between groups.

Figure 4.

AUROC for TPS (A), UTPS (B), and perceived rate (C) for all DDK tasks with associated Area Under the Curve (AUC) values. Dark black lines indicate those tasks for each measure which were most readily able to distinguish between groups of positive or negative concussion histories. Hashed lines are provided as the reference (.50) indicating no predictive value.

DISCUSSION

Objectively, the speech rate of athletes with SRC histories was slower during DDK tasks compared to controls. Athletes with SRC histories produced all DDK’s slower and with longer unvoiced time (pauses) per syllable compared to controls with no SRC histories. Concerningly, we identified that the speech rates of former nonprofessional athletes with supposedly resolved concussion symptomology, that is, no diagnosis of progressive disease related to repeated head injury, was clearly slowed when compared to healthy matched controls for simple speech DDK tasks.

Corroborating objective findings, eight trained SLP’s evaluated the speech of athletes with SRC histories to be slower than controls across DDK tasks. This indicates that even subtle changes in the speech rate may be detected by well-trained and qualified listeners. Finally, athletes with SRC demonstrated mostly increased lip and jaw muscle activation during the production of DDK tasks as compared to controls. Further, both objective and subjective measures of rate had good discriminatory power for athletes with SRC history as determined by AUROC analysis.

Objective Speech Rate Analysis

Normative DDK rate values have been established for healthy non-brain injured English speakers (ages 15-65) for the token puhtuhkuh at 6.2 syllables/second. Itch and Ben-David suggest that values of 5.4 syllables/second and slower may be potentially pathological and require further evaluation²¹. Healthy controls in our study produced 6.3 puhtuhkuh syllables/second (±.035) compared to 5.4 syllables/second (±.048) for those with SRC histories, a rate considered pathological²¹. Further, we found that DDK rate measures had fair to excellent discriminatory power with good sensitivity to reliably predict concussion history status. Given the minimal time and low cost of adding a DDK rate task to all concussion assessment time points, the high sensitivity and moderate specificity is acceptable. Further investigation will aim to increase the specificity of our measures.Importantly, we found high sensitivity and relatively low specificity in our analysis of TPS during “tuh” productions. This may be due to variability in the production of this syllable which may be resolved analyzing more data from larger samples. It is encouraging however, that for all other single and multi-syllabic productions, the TPS measure indicated significantly slower speech for those with a history of SRC.

Subjective Speech Rate Analysis

SLP perceptual evaluations of speech were consistent with the findings of the effect of SRC on objective timing analyses, where participants with SRC histories were rated as clearly slower during DDK repetitions than controls. Together with the objective timing evidence presented in this study, it is reasonable to deduce that participants with SRC histories did, in fact, produce clearly slower DDK repetitions. Additionally, SLP’s were reliably able to detect differences between groups with and without concussion histories, particularly during Buttercup repetitions. Thus, with proper training, speech professionals may be sensitive to speech differences in this population.

sEMG Analysis

Novel application of speech articulator muscle engagement analysis using sEMG provides additional evidence of the negative effects of SRC history on speech musculature function. Participants with SRC histories demonstrated increased muscle activation compared to controls on speech tasks routinely requiring the engagement and coordination of lip and jaw musculature. RMS-EMG results of articulatory speech muscle output described in this work are consistent with findings of increased cortical motor excitability due to increased neurotransmitter release and NMDA (excitatory neurotransmitter) receptor activation in individuals with SRC^36,37. This increase may be partially responsible for the decreased speech rate during DDKs tasks, resulting in increased rigidity and decreased articulatory function.

With the potential for SRC to cause widespread neurometabolic dysregulation, motor speech deficits may be related to cortical motor dysfunction akin to the early motor speech deficits indicative of ALS, Parkinson’s Disease and CTE^{5,8–10,38,39}. For example, the analysis of general speech rate decreases and pause increases, may be potentially sensitive in identifying those in the early stages ALS⁴⁰. Further, speech articulatory rate and pause measures have been used to identify individuals in early stages of Parkinson’s disease^38,41. Perhaps most intriguing are the implications speech motor evaluation may have for our understanding of CTE. CTE is characterized postmortem as degradation and toxic accumulation of pathogens in the cortical and subcortical structures responsible for cognitive and motor function. This degradation is likely multifactorial including repeated concussive and sub concussive head injury^10,11. To date, the studies that classify CTE speech patterns have not included younger, former, nonprofessional athletes with SRC histories, with no current overt symptom profiles, but instead, described older, former, professional athletes with suspected (self-reported) CTE who suffer from a characteristic disease and symptom profile.

By examining objective speech rate and timing, as well as the subjective judgements of articulatory precision, rate and rhythmic consistency, we have shown that speech of individuals with concussion history may not be classifiable as “slurred” as it relates to the only area of literature which has quantified the term (inebriated speech literature)⁴². We have shown in this work significantly increased pause times (unvoiced time) in those with a history of concussion. This is directly opposed to the operational definition and hallmark of slurred speech, the voicing and lengthening of normally unvoiced affricate speech sounds.

One limitation of our study was the variability in time following participants’ last SRC and being recorded for our study. While this was included in our statistical model and did not show a significant main effect, uniformity in this factor would be ideal to produce more generalizable results. Additionally, future work should control for participants’ age at first injury as this may influence speech production outcomes. Future work in this area should incorporate structural and functional neuroimaging to inform therapy and potentially increase motor speech ability in this population. Finally, our sample did not include athletes with acute SRC and active symptoms. Given that participants were no longer experiencing SRC related symptoms, we anticipate that analyses presented in this work may be used to identify individuals with acute (active symptomology) and non-acute SRC, persistent concussive syndrome, and the potential severity of SRC injuries. This limitation is one that should be addressed in future work by collecting speech samples from athletes with acute concussion, with and without ongoing symptoms. We anticipate that athletes with acute SRC would demonstrate similar or worse speech rate and muscle activation outcomes during acute stages of SRC. Future work will apply the motor speech analysis methods outlined in this study to predict and track SRC recovery, degenerative disease, and potential risk factors associated with long-term outcomes.

CONCLUSIONS

SRC history may be associated with long-lasting deficits in objective speech timing, physiological articulatory muscle power, and subjective evaluations of rate exist between individuals with SRC histories compared to those without. Our findings have clinical implications as measures established in this work may have diagnostic and prognostic utility, be used in a mobile fashion, and be sensitive to changes over time which can inform return to play recommendations.

What does this study add?

This study is the first of its kind to analyze speech timing and acoustic measures of athletes with and without SRC histories to demonstrate differences between groups.

Biographies

Biographical Notes

REB: Dr. Banks is a postdoctoral fellow at Massachusetts General Hospital in the Brain Recovery Lab (Dr. Teresa Kimberley).

DSB: Dr. Beal is an assistant professor at the University of Toronto Speech-Language Department and a clinician scientist in the Bloorview Research Institute - Holland Bloorview Kids Rehab Hospital.

EJH: Dr. Hunter is a foundation professor, associate dean, and director of the Trifecta Initiative at Michigan State University in the Communicative Sciences and Disorders Department.

Data Availability

Data for this project is available from the primary author given reasonable request.