Listener Detection of Objectively Validated Acoustic Features of Speech in Huntington’s Disease

Subjects

Sixty participants were recruited (30 people with HD CAG expansion and 30 age-matched healthy controls). Potential controls were excluded if they reported a history of neurological, speech or language disorder. Potential HD participants were excluded if they presented with: any other neurological disease other than HD; clinical symptoms other than that resulting from HD; a history of communication disorder; a history of alcohol or drug abuse; or a history of learning disability and/or intellectual impairment. Two groups of participants with the HD CAG expansion (16 people with PreHD and 14 people with EarlyHD) were recruited from Monash University Huntington’s Disease Registry, St George’s Health Service Huntington’s Disease Clinic and Huntington’s Victoria, Australia. Due to differences in average age in the PreHD and EarlyHD groups, from our overall control sample, we divided controls into two age-matched groups (Control-A and Control-B). We computed disease burden scores (DBS) using a formula (age x [CAG repeat –35.5]) for CAG-expanded participants, which correlates pathological progression and striatal damage [23] (see Table 1 for clinical characteristics of HD participants).

The PreHD group consisted of people who were genetically confirmed to have the HD CAG expansion (39 repeats or more), but had not yet displayed motor signs sufficient to warrant a clinical diagnosis of HD. Diagnostic confidence levels (DCLs) are based on administration of the United Huntington’s Disease Rating Scale (UHDRS) and are aligned with the confidence a clinician has in diagnosing HD in a particular participant (0 = no abnormalities; 1 = non-specific motor abnormalities; 2 = motor abnormalities that may be signs of HD (50–89% confidence); 3 = motor abnormalities that are likely signs of HD (90–98% confidence); and 4 = motor abnormalities that are unequivocal signs of HD) [21]. Classification of PreHD was based on DCL≤3. The PreHD group was age-matched to 16 healthy participants without neurological disorders in Control-A group (age range 25 to 72 years, mean = 42.38, SD = 12.09, male = 44%) for comparison (Table 1). PreHD and Control-A groups were similar in age (p = 0.98). The EarlyHD group included 14 symptomatic participants with DCL of 4 as assessed by a neurologist specialized in movement disorders. EarlyHD group was age-matched to 14 neurologically healthy individuals in Control-B group (Control-B: age range 41 to 71, mean age = 56.29, SD = 10.32, male = 64%) (Table 1). The EarlyHD and Control-B groups were also similar in age (p = 0.95). The International Standard Classification of Education (ISCED) was used to compare educational levels and related qualification levels between participant groups (PreHD: mean = 4.08, SE = 0.31; EarlyHD: mean = 3.44, SE = 0.33; Control: mean = 4.36, SE = 0.25) [24]. There was no significant difference on ISCED mean values be-tween participant groups.

Materials and stimuli

A formalized speech battery was administered with each participant individually in a quiet setting. All participants were asked to (i) sustain a vowel sound /a/, (ii) read aloud a phonetically balanced passage –The Grandfather passage [25], and (iii) produce a monologue. The order of the speech tasks administered was consistent across all participants. The three speech tasks and corresponding acoustic measures have known stability, reliability and sensitivity with repeated exposure and application of the stimuli (i.e., practice effects) [26]. Speech samples were recorded using a laptop computer (PC) (Hewlett-Packard, Palo Alto, CA) with basic factory settings and a Sennheiser PC 135 USB unidirectional head-mounted microphone (Sennheiser Communications, Solrød Strand, Denmark) (minimum sensitivity of –38 dB and a frequency range of 80 Hz–15 kHz), which was positioned at a 45 angle, 8 cm from the mouth [27]. All the data were sampled at 44.1 kHz, coupled with quantization at 16 bits. Data were recorded and segmented using Audacity software (version 1.2.6). In addition to perceptual evaluation of speech and voice, we used an automated script to conduct acoustic analysis, including speech-timing measures for the reading and monologue tasks (See Table 2 for Descriptions of speech-timing measures). The automated script identified the silence intensity contour using a modified version of techniques published previously [28, 29]. Pause sections that were shorter than 15 ms were classified as speech and concatenated with the adjacent speech sections. Speech sections that were shorter than 30 ms were classed as pauses and concatenated with the adjacent pauses [29]. We also removed interjections (i.e., throat clearing, fillers) as part of the pre-processing procedure of speech samples. Detailed explanation of the specific acoustic methodologies adopted in this study can be sourced from previous published papers [26, 28–30].

Statistical analysis

The degree of agreement was calculated between raters using a Two-Way Random Consistency intraclass correlation coefficient (ICC 2, 1). Listener-based perceptual speech ratings were analyzed by a nonparametric test of gamma coefficient, which investigated the link between groups (PreHD, EarlyHD, Control-A, Control-B) and judgements of speech impairment severity (normal, sub-clinical, mild, moderate, severe). We used the Spearman Rank method to examine the degree of association between perceptual ratings and acoustic outcomes.

For the statistical analysis of acoustic data, we used the Levene’s test to assess equality of variances between groups in their acoustic speech outcomes. Results from the Levene’s test revealed equal variances between groups. A one-way analysis of variance (ANOVA) was then used to evaluate speech-timing differences between the three independent groups (PreHD vs. Control-A, EarlyHD vs. Control-B, PreHD vs. EarlyHD). All statistical analyses were performed using Statistical Package for the Social Science (SPSS Statistics 26).

Auditory-perceptual profile of speech in PreHD and EarlyHD

Agreement between raters was 72.2% (ICC =0.72). Perceptually, speakers in the PreHD group exhibited abnormal pitch level (p = 0.007), reduced loudness (p = 0.007), dysphonic voice quality (p < 0.05), hypernasality (p = 0.007), abnormal speech rate (p < 0.001), imprecise articulation (p = 0.007), and reduced naturalness of speech (p = 0.004) compared to their age-matched control group (Table 3). Most of these speech deviations were perceived at the sub-clinical level rather than clinically dysarthric (i.e., mild to severe). Speakers in the EarlyHD group demonstrated mild-to-moderate dysarthric features in their speech, including unchanged or gradual reduction in loudness (p < 0.01), dysphonic voice quality such as harsh, hoarse, breathy, and strained voice quality (p < 0.05), abnormal prosody (p < 0.01), and articulatory breakdowns and speech-sound distortions (p≤0.001) compared to their matched-control group (Table 3). Overall, speech in EarlyHD was less intelligible (p = 0.005) and less natural (p < 0.001) compared to PreHD and healthy control groups (see Fig. 1). Table 3 shows statistical comparison outcomes and Supplementary Table 1 presents raw data on listener-based speech rating.

Fig. 1

Frequency of speech deficits (in percentages) across PreHD, EarlyHD, and control participants based on perceptual assessment. ^aParticipants from Control-A and Control-B were combined in Fig. 1 (n = 30). Significant difference between PreHD and control groups: ^*p < 0.05; ^**p≤0.01; ^***p≤0.001. Significant difference between EarlyHD and control groups: ^*p < 0.05; ^**p≤0.01; ^***p≤0.001.

Computational acoustic analysis indicated the rate of speech on reading tasks significantly differed between the PreHD group and healthy control group. The effect of group on speech rate in PreHD suggested that the reduced rate of speech was related to disease stage (p < 0.001) (Table 4). In addition to reduced speech rates, people with EarlyHD demonstrated longer pauses and a higher proportion of silence compared to the PreHD and control groups on reading and monologue tasks (p > 0.05) (Table 4). PreHD and control groups performed better than EarlyHD group on oral SDMT (p < 0.001). Faster speech rate on the reading task correlated with higher oral SDMT score (PreHD: r = 0.72, p = 0.01; EarlyHD: r = 0.54, p = 0.04). The relationships between perceptual and acoustic features of speech-timing, and between SDMT and acoustic speech outcomes are described in Supplementary Tables 2 and 3.

The link between cognition and speech-timing in HD

Slower speech rate in participants with the HD gene expansion is associated with a common cognitive measure known for its sensitivity in HD, the SDMT, suggesting that speech rate declines in parallel with cognitive decline in HD. Another study investigating people with multiple sclerosis have also reported similar association between cognitive functions and variances in speech-timing using the Minimal Assessment of Cognitive Function [35]. Slower speech rate is also observed in prodromal idiopathic rapid eye movement (REM) sleep behavior disorder, which is associated with a higher risk of cognitive impairment [36]. In the HD population, PreHD can manifest in deficits of executive function, involving working memory and cognitive control [19, 37, 38]. Data from the PREDICT-HD study suggests that several cognitive measures track disease progression in PreHD [37]. Speech production involves various cognitive-linguistic processes such as lexical (word) processing, syntax system (grammar) processing, and phonemes (speech sounds) encoding [39, 40]. Cognitive decline, specifically cognitive processes theorized to be involved in linguistic planning and formulation such as verbal working memory and attention [41–43], may negatively compromise speech-timing performance in speakers with PreHD and EarlyHD.

Limitations

Future research is needed in HD to distinguish how speech production relates independently to the motor and cognitive changes occurring in HD. The procedure for eliciting responses in the oral SDMT presents a challenge for individuals with a real or potential speech impairment, as they are required to respond to a timed test orally which in turn is likely to influence their performance [44, 45]. Given that performance on oral cognitive tasks is thought to be independent from oral motor control, the precise impact of cognition versus oral motor execution on cognitive task performance requires further delineation. Future studies should mitigate the impact of oral motor control in neuropsychological testing, as well as designing a more comprehensive test battery to measure specific cognitive domains other than attention and processing speed from the oral SDMT, such as executive function. Other cognitive tools that include both accuracy and speed of performance might be beneficial, accuracy of performance should not be impacted by oral motor or fine motor control in upper limb movement. Future studies can also evaluate the relationship between speech and cognition in healthy controls which could present differently compared to the HD population.

Another limitation of the study relates to the potential effect of mental health disorders on speech rate. Depression is one of the most common psychiatric symptoms in HD and it can occur in the premanifest phase of HD [46–48]. Individuals with depression and anxiety have shown to display slower speech rate and increased pauses [49], yet prevalence of depression was not measured in this study.