Participants

All participants provided written consent in compliance with the Boston University Institutional Review Board under protocol 2625, which approved this study. Twenty-one speakers with PD and twenty-eight control speakers were recruited to participate in the study. Four speakers with PD were excluded from the study group due to abnormal hearing thresholds for older adults (N = 2, see Hearing Thresholds) or inability to complete the main experimental task (N = 2). Eight control speakers were excluded from the study group due abnormal hearing thresholds (N = 6), perceptually atypical voice (N = 1), or failure of a pre-study cognitive screening (N = 1). One speaker with PD and one control speaker were excluded due to data analyses restrictions on the primary experimental task (see Data Analysis). Thus, sixteen speakers with PD aged 52–73 years (7 female) and nineteen healthy control speakers aged 50–77 years (8 female) were included in the study (see Table 1). The complete study lasted two sessions. Fourteen speakers with PD and eleven control speakers completed both sessions, and two speakers with PD and eight control speakers completed a single session. All speakers with PD were diagnosed with idiopathic PD by a neurologist prior to participation in the study and were receiving daily levodopa/carbidopa therapy. Twelve naïve listeners aged 18–23 years (6 female) were included in the study to assess the speech intelligibility of the speakers with PD (see Data Analysis). Control speakers and listeners denied any known neurological, speech, hearing, cognitive, or language disorders.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Participant characteristics (M: mean, SD: standard deviation).

https://doi.org/10.1371/journal.pone.0191839.t001

Movement Disorder Society Unified Parkinson’s Disease Rating Scale (MDS-UPDRS)

Part III (Motor Examination) of the Movement Disorder Society sponsored revision of the Unified Parkinson’s Disease Ratings Scale [MDS-UPDRS; 40] was provided to each participant with PD to assess the motor signs of PD. This documents the severity of abnormalities during execution of motor function tasks such as speech, upper and lower extremity movements, and walking. The first author, certified to administer MDS-UPDRS, scored each examination per protocol. The UPDRS total motor score (see Table 1) was used as a correlate of disease progression.

Hearing thresholds

Each participant underwent pure-tone hearing threshold testing at 250, 500, 1000, 1500, 2000, 3000, and 4000 Hz using 3M^TM E-A-RTONE^TM Gold 3A insert earphones and the Grason-Stadler GSI 18 Screening Audiometer. All participants had hearing thresholds within normal range for older adults [under 25 dB HL for frequencies 1000 Hz and below, and under 40 dB HL above 1000 Hz; 41]. None of the participants wore hearing aids.

f_o Adaptation procedure

All experimental sessions were conducted in a sound-attenuating audiometric booth. Participants were seated in front of a computer monitor and wore a Shure omnidirectional MX153 earset microphone positioned at approximately 45 degrees from the midline and 7 cm from the corner of the mouth. The microphone signal was amplified via an RME Quadmic II microphone preamplifier and digitized via a MOTU Ultralite-mk3 Hybrid soundcard with a sampling rate of 48000 Hz. The auditory feedback was amplified at least 5 dB relative to the sound pressure level (SPL) at the microphone [based on relative dB between microphone and ear; 42]. Due to the length of the experimental procedure (3 conditions × 30 minutes each), sound exposure was kept to the minimum possible sound pressure level that allowed participants to adapt to auditory feedback changes. For calibration, the intensity of the auditory feedback in the headphones was measured using an IEC 60318–1 coupler (Type 4153, Bruel & Kjaer Inc., Norcross, GA) connected to a sound level meter (Type 2250 Hand Held Analyzer with Type 4947 ½” Pressure Field Microphone, Bruel & Kjaer Inc., Norcross, GA) such that an electrolarynx (TruTone Artificial Larynx, Griffin Laboratories, Temecula, CA) input with an intensity of 75 dB SPL approximately 7 cm from the microphone resulted in at least 80 dB SPL output in the headphones. Auditory feedback was administered via Sennheiser HD 280 Pro headphones, which attenuate air-conducted sound by approximately 32 dB (Sennheiser Electronic Corporation). The Sennheiser HD 280 Pro headphone set used across participants had a different response for some PD and control participants, but data were analyzed for adaptation response differences and there was no clear effect of the headphone response. The f_o was manipulated using Audapter [43], a MATLAB software package for configurable real-time manipulation of acoustic parameters of speech, using the ASIO4ALL driver. The total processing delay was 45 ms, which is below the threshold at which delayed auditory feedback can cause speech dysfluency [44].

The adaptive responses to f_o perturbation were recorded under three conditions: ‘‘shift-up”, “shift-down”, and “control”, with the order counterbalanced across participants. Each condition consisted of 160 trials and each trial was 11 s in duration. Participants were instructed to produce a sustained /α/ for 3 s when “aaa” was displayed in green on the computer monitor and to then rest their voice for the remaining 8 s until the next trial began. The “shift-up” and “shift-down” conditions consisted of trials over four ordered stages: baseline, ramp, hold, and after-effect. In the “shift-up” condition, the first 20 utterances, referred to as the baseline, were produced while receiving typical (unperturbed) feedback. In the following 60 trials, referred to as the ramp, the f_o in the auditory feedback increased by 1.69 cents with each successive trial, reaching a total level of 100 cents of perturbation above the participant’s true f_o. For the next 40 trials, referred to as the hold, the f_o of the auditory feedback was maintained at the level of +100 cents of perturbation. In the last 40 trials, referred to as the after-effect, the auditory feedback was the same as the baseline (i.e., there was no f_o perturbation). The “shift-down” condition was identical, except that, during the ramp, the f_o of the auditory feedback decreased by 1.6 cents in each successive utterance, reaching a total level of -100 cents of perturbation by the end of the ramp. Perturbations during the ramp and hold phases were applied throughout the entire period of voicing. In the “control” condition, participants received their typical (unperturbed) feedback during all 160 trials (similar to the baseline and after-effect periods of the “shift-up” and “shift-down” conditions). The participants did not receive any information about differences among the three conditions.

Pitch acuity

All participants completed a pitch discrimination task to assess auditory acuity to f_o, or pitch acuity. Each trial consisted of two pure tones, presented via Sennheiser HD 280 Pro headphones. Each tone was 2 seconds in duration. The tones in each trial were separated by approximately 20 ms and were presented in random order. Stimulus presentation was controlled using a custom-written MATLAB (The Mathworks, Inc., 2013, Version 8.1.0.604 [R2013b]) script. Tones were played at an intensity of 65 dB, calibrated using the same method as the adaptation procedure but using the MATLAB generated stimuli as the input. Participants were asked to judge whether the tones were the ‘same’ or ‘different’. The initial difference between the two tones was 40 cents (10 Hz) with a reference tone of 440 Hz. Two consecutive correct judgments resulted in a progressively smaller difference (downstep) between the two tones in subsequent trials, whereas an incorrect judgment resulted in a progressively larger difference (upstep) between subsequent sets of tones, with a set ratio of downstep = upstep/2.448. The magnitude of the upstep was initially 20 cents (5 Hz). As trial number increased, the upstep decreased to 10 cents (2.5 Hz), 5 cents (1.27 Hz), and 1 cent (0.25 Hz), at the tenth, twentieth, and thirtieth trial, respectively. The experiment continued until there were 16 ‘reversals’ (an upstep followed by a downstep, or vice versa). Catch trials with no difference between the two tones were presented pseudo-randomly in 33% of trials to ensure data validity and attention to the task. Each participant performed between 36 and 85 trials, with an average of 70 trials. The duration of this task was 6–10 min.

Speech Intelligibility Test (SIT)

Speakers with PD completed a randomly generated short Speech Intelligibility Test [SIT; 45] consisting of a set of 11 sentences increasing in word count from 5 to 15 words per sentence. Speakers were instructed to read the sentences in their typical speaking voice. Using the same recording equipment as the adaptation paradigm, speech was recorded in Audacity (Audacity Team, 2012, Audacity®. Version 2.0.0. audio editor and recorder).

Data analysis

Adaptive responses to f_o perturbation in PD

Sensorimotor adaptation has been studied extensively in PD, but only one prior study investigated sensorimotor adaptation in PD in the speech domain [31]. Adaptive responses to perturbations of the first formant of vowels (F1) during word productions were analyzed and reported at a group level. The study found reduced responses in the PD group compared to the healthy control group [31]. The data presented here, at a group level, extend this finding to the control of f_o. Furthermore, on an individual level, these data demonstrate qualitatively different responses between control speakers and speakers with PD. In speakers with PD, responses were also not consistent among shift directions (ie, speakers who compensated in the shift down condition did not always compensate in the shift up condition or vice versa). While control speakers showed relatively consistent compensatory responses to perturbation, speakers with PD had highly variable responses, with either compensatory responses, no response, or even “following” responses (changing f_o in the same direction as the perturbation). There was no clear pattern in these responses. In the PD group, 2/15 subjects only completed one shift condition. Of the remaining 13/15 who completed both conditions, 7/13 had the same result regardless of direction (3 compensated, 2 did not respond, and 2 followed). The remaining 6/13 subjects who completed shifts in both directions did not show the same response in the two directions: 2 compensated in one direction and did not respond in the other, 3 compensated in one direction and followed in the other, and 1 followed in one direction and did not respond in the other. Using a reflexive paradigm, Kiran and Larson [34] also found more “following” reflexive responses to unanticipated perturbations of f_o in speakers with PD than in control speakers. However, the relationship between following reflexive and following adaptive response types in PD is unclear, especially given the lack of consensus between adaptive and reflexive response types in general.

While the group-level responses in the current findings align with those reported by Mollaei, Shiller [31], the individual responses differ. The heterogeneous results seen in the current work were an unexpected outcome of the study. These responses could be due to several differences between the current work and Mollaei, Shiller [31]: perturbing vocal vs. articulatory characteristics of speech (f_o vs. F1) or differences in how the perturbation was introduced and the experimental set up. Perturbations of f_o and F1 may involve different control systems. Based on short- and long-term effects of auditory deprivation, it has been hypothesized that so-called ‘postural’ parameters such as f_o are more heavily reliant on auditory feedback control processes than so-called ‘segmental’ parameters such as F1, which are more heavily reliant on feedforward control processes [51, 52]. This suggests that a potential feedforward system impairment in auditory-motor integration [refinement and updating of the feedforward system basedon on systematic changes in auditory feedback; e.g., 21] might have a greater effect on articulatory (F1) motor control relative to vocal (f_o) motor control. Furthermore, reduced loudness is a hallmark feature in PD [4], which has a biomechanical relationship with f_o [53, 54]. Thus, some interaction between hypophonia and f_o in the speakers with PD could have contributed to the variability in f_o adaptive responses. Other than the parameter to which the perturbation was applied, there are additional methodological differences between the current study and Mollaei, Shiller [31]. The paradigm used by Mollaei, Shiller [31] differed from most current speech adaptation paradigms [15, 29, 55]: a ramp phase was not included to gradually introduce the perturbation. This paradigm difference is important to note when comparing the adaptive responses. Visuomotor studies have shown differences in responses to sudden (no ramp) and gradual introduction (with ramp) of perturbation in healthy controls [56, 57] and individuals with PD [58]. Lastly, the current study applied auditory perturbations in cents, which are a relative measure of frequency, rather than changing feedback by the same set frequency for all subjects. This is a common method used in f_o perturbation paradigms [16, 25, 30, 59] and ensures that all subjects perceive the same magnitude of feedback perturbation since the just-noticeable-difference for frequency acuity increases as a function of frequency [60, 61]. In formant perturbations, shifts are applied as a set frequency change in the vowel formants rather than a subject specific shift in cents [27, 62], which may contribute to variations in results between the current work and formant perturbations in PD [31].

The individual responses seen in the current study during the after-effect stage align well with a study of visuomotor control in PD. Contreras-Vidal and Buch [63] found heterogeneous responses in speakers with PD during the after-effect stage of a visuomotor adaptation task that were not seen in healthy controls. Similar heterogeneous responses were found in the current work during the after-effect stage in the PD group; however, these may have been a continuation of the heterogeneous responses that were also seen in the ramp and hold.

In terms of the relationship between baseline f_o variability and degree of compensation, the current results showed no significant correlations for either speaker group for both “shift-up” or “shift-down” conditions (Table 2). The results for speakers with PD are in contrast to Huang, Chen [47] who reported a significant positive correlation in a PD group between baseline f_o variability (during a 200 ms baseline phase) and the degree of compensation to unexpected f_o perturbations. The lack of significant correlations in the current work is likely due to the differences between so-called “reflexive” and “adaptive” responses to sensory perturbations. Such distinctions between unexpected reflexive shifts and studies of adaptation (repeated perturbation applied on many consecutive trials, causing changes in the sensorimotor control system that continue to affect movements for a number of trials after the perturbation has been removed) have been made in a large number of studies of both speech production [e.g. 15, 64–66] and motor control more generally [e.g. 56, 57, 58, 67–70].

Interpretation of potential neurophysiological mechanisms

The atypical responses to gradual, repeated, f_o perturbation seen in the current study support the hypothesis that there may be an impairment in the auditory-motor learning process in PD. Using the DIVA and SFC models to interpret these findings, in neurotypical participants, the feedback system compares the output f_o to the target f_o, and generates an error signal due to a mismatch between the two. In the case of the SFC model, an internal vocal tract model is also present and is continuously keeping track of the state of vocal articulators to use this information for future commands. If either the feedback controller or the internal model (in the case of SFC) register an error signal, a corrective motor command is generated to oppose the shift in f_o. This is communicated to the feedforward subsystem which is adapted (over multiple perturbed productions) to reduce the mismatch between target and output in future productions. The heterogeneous responses seen in the adaptive responses of the individuals with PD suggest that either: 1) the f_o perturbation was not correctly detected by the auditory feedback controller (DIVA) or the internal vocal tract model (SFC), 2) detected auditory errors were not properly translated into corrective motor commands, or 3) the feedforward system did not correctly modify motor commands for subsequent productions by incorporating corrective motor commands from the auditory feedback controller. These possibilities are addressed in the following paragraphs.

If atypical adaptive responses were due to a problem with f_o detection in the auditory system in speakers with PD, this would suggest less sensitive pitch discrimination compared to healthy speakers. In the current results, there was no significant difference in pitch acuity between speakers with PD and control speakers. Additionally, pitch acuity was not correlated with the degree of compensation in adaptive responses in speakers with PD. Interpreting these results with the DIVA and SFC models, this indicates that there was not likely a deficit in detecting pitch changes in the speakers with PD. The pitch acuity findings in the current study are in contrast with previous work that found reduced ability to discriminate pitch in 12 individuals with PD compared to 15 healthy controls [71]. The conflicting results could potentially be explained by natural variation in musical experience of participants, paradigm differences, or the limited sample sizes employed by both studies. Previous work has shown that musicians may be better than non-musicians at discriminating pitch [72, 73]. Since neither Troche, Troche [71] or this current study reported musical training, it is possible that it contributed to differences between study results. Paradigm differences between Troche, Troche [71] and the current study could also explain the different findings. Among other differences, tones were presented with an inter-stimulus interval (ISI) of 750 ms in Troche, Troche [71]. In the current study, an ISI of only 20 ms was used. It is possible that the longer ISI employed by Troche, Troche [71] is responsible for the poorer pitch acuity seen in speakers with PD in that study: assessment procedures for frequency discrimination are known to be impacted by relative attentional and cognitive ability [74]. In fact, reduced cognitive function has specifically been shown to interact with ISI, resulting in differentially poorer frequency discrimination in individuals with Alzheimer’s disease when long ISIs are used [75]. Future studies comparing pitch acuity measures as a function of ISI in speakers with and without PD are necessary to confirm this. Finally, especially given the heterogeneity in PD presentation, differences between the two studies may simply be a function of sample size. That said, since pitch acuity was not correlated with the adaptive responses in the speakers with PD in the current study, it does not appear to be a driving factor of the adaptive responses in this sample. However, given that individuals with PD demonstrate difficulty in correcting for sensorimotor errors during self-generated movements [76–78], we cannot rule out the possibility that the perception during self-productions (autophonic judgments) may have contributed to the atypical adaptive responses in PD. Previous work has found that when listening to their own voice playback, individuals with PD are able to determine the vocal loudness within normal limits [79], whereas during an active speech task, individuals with PD overestimated their own vocal loudness [80]. Additionally, Huang, Chen [47] revealed differences at the cortical level between individuals with PD and healthy controls during an active speech task with frequency-altered-feedback, but no differences were seen when individuals with PD were passively listening to the playback of their own voice.

The atypical responses seen in the current study could also be due to a deficit in the feedback system in correcting the detected errors. However, because studies of reflexive responses to sudden, brief, perturbation of f_o have demonstrated that speakers with PD have larger compensatory response magnitudes [30, 32, 33] relative to control speakers, there is evidence that the ability to produce corrective responses to detected errors during auditory perturbations is not impaired in PD. Thus, it is unlikely that the current results are due to an inability of the feedback system to compensate for detected errors in f_o. The larger reflexive compensations in the PD group seen by Liu, Wang [33] were in response to a 100 cent f_o perturbation and Chen, Zhu [32] reported similar results with 50, 100 and 200 cent perturbation conditions. Additionally, Chen, Zhu [32] found an increase in response magnitude as perturbation size increased. Mollaei, Shiller [30] also observed that a PD group had larger compensatory response gain as compared to controls when the perturbation was 50 cents, but found no group difference in compensatory response gain when the perturbation was 100 cents. Larger compensation magnitudes suggest that the feedback system is not impaired and that, in fact, there may be a higher reliance on the feedback system in PD than in healthy speakers during reflexive responses to f_o. It is possible that speakers with PD rely more on their auditory feedback for f_o control because of a feedforward system impairment. Additionally, these larger compensations may be due to a deficiency in the weighting of somatosensory and auditory feedback in PD. This weighting in known to vary in typical speakers, but more extreme variation could explain the results shown here [28].

The current results, together with previous work, most directly support the possibility that f_o control in speakers with PD is impaired due to a deficit in the updating of the feedforward control system after an error signal has been correctly generated. This would mean the f_o changes are accurately detected and corrected by the feedback control system in speakers with PD, but these corrective responses are not properly subsumed into the feedforward system. This interpretation aligns with previous work suggesting the basal ganglia have a role in generating feedforward commands [81–83] and that the striatum (the primary target of cortical projections to the basal ganglia) contributes to motor learning processes [84]. Therefore, the reduced dopaminergic signals in the striatum seen in PD [85, 86] would impact the integration of corrective motor commands into consequent motor actions. This is further supported by studies which found atypical sensorimotor integration [63, 87–91], including auditory-motor integration [31], in PD. To clarify the contributions of the feedback and feedforward system to vocal motor control in speakers with PD, a comprehensive experiment with both unexpected and sustained perturbations of f_o is necessary.

Limitations and future work

A limitation of the current study, as in most studies in PD, is the heterogeneity in the presentation of the speakers with PD in terms of the timing of the onset of their symptoms and the severity of their symptoms. Individuals with PD reported with a wide range in the time between the onset of PD symptoms and their participation in this study (1.5–30 years), as well as in their UPDRS total motor score (ranging from 22–72). The variability in disease severity is difficult to circumvent in studies of PD, as there are a wide range of motor and non-motor symptoms seen in the clinical population of individuals with PD [92–95]. However, progression of dysprosody in PD does not appear to correlate with disease duration or severity as assessed by the UPDRS motor score [96, 97].

Medication state may also have influenced the current study. All speakers with PD were tested while they were receiving L-dopa therapy which allows for variability in duration of administration as well as the individual’s symptoms at the time L-dopa therapy was initiated, influencing the effects of the medication [98–102]. Previous research on speech production in PD has not demonstrated substantial differences in mean f_o [103–106] or f_o variability [97, 105, 107] between participants on and off medication. However, during brief perturbations of f_o, larger responses were reported in speakers with PD off medication as compared to a healthy control group [30, 32, 33], whereas one study [34] found that ten speakers with PD on medication did not differ from ten control speakers in terms of reflexive response magnitude. Speakers with PD were not directly compared on and off medication in these studies, but the combined results imply there could be an effect of medication on auditory-motor control. In summary, the effects of medication state on adaptive responses in the auditory domain are not defined, especially given the heterogeneity of responses to medication in the PD population, and should be further investigated by examining adaptive responses of participants who are both on and off medication. The perception of reduced fluctuation in voice fundamental frequency, or ‘monopitch’, observed in PD may also have affected the PD group’s ability to change their f_o. However, we would expect such an effect to show reduced or absent responses while the PD group, on an individual level, showed responses with magnitudes comparable to the control group. Another factor that could have contributed to variation in the current study is the cognitive function of individuals with PD. Previous studies have shown that as attentional demands increase, control speakers have reduced responses to both sudden f_o perturbation [108] and gradual, predictable, f_o perturbations [109]. Many individuals with PD have cognitive deficits, particularly in attention [110–112]. However, attention deficits in PD appear to mainly affect dual-attention or divided attention tasks [113, 114], and therefore were most likely not a concern for the current study because it was a single task experiment.

Lastly, there was a 45 ms delay in auditory feedback during the experimental procedure. To our knowledge, only one study has examined the impact of delay on adaptive responses to auditory perturbations. Max and Maffett [115] conducted a formant adaptation task under various delay conditions: no delay, 100 ms, 250 ms, and 500 ms delay. Results showed auditory-motor adaptation was eliminated in all delay conditions. Unfortunately, they did not examine delays between 0 and 100 ms. The healthy controls in the current study showed clear adaptive responses and did not exhibit any speech dysfluencies, suggesting auditory-motor adaptation was not affected by the 45 ms experimental delay. However, it is important to note that the impact of feedback delay on speakers with PD compared to control speakers is unclear and requires further investigation.