Vowel Production Associated with Cantonese Speakers of Parkinson’s Disease

Abstract

Introduction

Patients of Parkinson’s disease (PD) often show symptoms of hypokinetic dysarthria and vowel distortion. However, the exact change in vowel production and its relationship with disease progression is still not clear. The present study examined the vowel characteristics associated with PD individuals of different severities.

Methods

The first two formants (F1 and F2) associated with the vowels (/i, u, a/) produced by 18 (11 mild and 7 moderate) PD patients and 30 healthy controls were obtained. Based on these, six derived formant metrics were calculated and used to depict distribution of vowels.

Results

Significant differences were found in all derived formant metrics between healthy and PD speakers, and there was a trend that the differences increased with severity of PD. Speakers of mild PD also exhibited a significantly smaller F1RR and tVSA than healthy speakers, and a greater F1RR than speakers of moderate PD.

Conclusion

The present findings suggest a reduced working vowel space with an apparent vowel centralization in PD speakers when compared with healthy controls. This was accompanied by restricted jaw and tongue movements. Such findings confirmed the articulatory undershooting in PD speakers, possibly due to hypokinetic dysarthria and other PD symptoms such as muscle rigidity and bradykinesia. It is also suggested that tVSA and F1RR are sensitive formant metrics for subtle changes in vowel articulation in PD.

Plain Language Summary

This study investigated how Parkinson’s disease affects the ability to produce clear vowel sounds among Cantonese patients. It is found that even in early stages, patients have a smaller vowel space, implying that their tongue movements are more limited, making their vowels sound distorted, as suggested by the various acoustic measures. This finding suggests that speech therapists should focus on improving articulation. The study also indicates that specific measurements of vowel sounds could be used as sensitive markers to detect these subtle speech changes early on, aiding in better management of the condition.

Introduction

Parkinson’s disease (PD) is a neurodegenerative disorder with a global prevalence rate of 0.1–0.2%, equating to more than 10 million PD patients worldwide [1, 2]. The prevalence of PD has been reported to increase with age. Tysnes and Storstein [3] reported that PD is affecting 1% of the population above 60 years of age worldwide. A similar prevalence rate of 1.06% was reported for China in a systematic analysis [4]. According to the Parkinson’s Foundation of the USA, there are about one million individuals in the USA are living with PD, a number projected to increase to 1.2 million in 2030 [1]. Between 1990 and 2019, the estimated prevalence of PD rose by 0.61%, predominantly among individuals aged 80 or above. The USA and Norway witnessed the most substantial increases, with rates up to 2.87% and 2.14%, respectively [5].

In addition to disruptions in movement and posture, many PD patients experience various speech-related difficulties. About 80% of PD patients exhibit hypokinetic dysarthria, a speech disorder that tends to affect all aspects of speech production including respiration, phonation, articulation, resonance, and prosody [6]. PD medication, such as levodopa, can generally alleviate many motoric symptoms, such as rigidity, stiffness, and bradykinesia. Yet, dysarthria has been reported to worsen over the course of disease progression despite the medication, leading to a progressive deterioration in speech intelligibility and effective communication. Consequently, patients often face increased social isolation and a diminished quality of life [7]. In fact, many PD individuals and their caregivers reported that such speech disturbances constitute the “most difficult aspects” of PD patients [8].

The speech deficits observed among PD patients are a direct manifestation of PD’s underlying pathophysiology. At the articulatory level, rigidity and bradykinesia reduce the speed and range of motion of speech articulators (jaw, tongue, lips), leading to articulatory undershooting during vowel production, where movements fail to reach their intended phonetic targets. As a result, the acoustics associated with vowels, particularly vowel formants, are disrupted. The theoretical pathway from nigrostriatal dopamine depletion, basal ganglia dysfunction, motor deficits, to articulatory impairment, and finally acoustic changes forms the foundation for investigating formant metrics as sensitive markers for the severity of speech-specific motor compromise in PD.

Imprecise vowel articulation greatly contributes to the decline in speech intelligibility, even in cases of mild PD [7, 9]. Kinematic data have revealed a general articulatory undershooting in PD individuals, accompanied by reduced loudness and speed of articulatory movement, including those of the jaw, lips, and tongue [10]. These issues likely are resulted from rigidity, bradykinesia, and hypokinesia of speech articulators, which are predominant characteristics of PD-related speech problems [11]. Acoustic studies of vowels produced by PD speakers, many of which relied on vowel space area (VSA) measures, have provided a general understanding of altered vowel distribution in PD speakers. However, discrepant findings have been reported [12]. Tjaden et al. [13] reported diminished vowel dispersion and reduced VSA in individuals with PD. In contrast, other studies found no significant difference in VSA between PD and healthy speakers [9, 10, 14]. These discrepancies have led some researchers to question if VSA is sensitive enough to accurately describe changes in vowel production in PD, particularly for those with mild to moderate dysarthria [9, 15, 16].

Subsequently, other derived formant metrics were developed, including (1) triangular VSA (tVSA), (2) vowel articulation index (VAI), (3) formant centralization ratio (FCR), (4) average vowel spacing (AVS), and (5) formant range ratios. Calculated based on three corner vowels of a language (/i, a, u/), tVSA refers to the area of the triangle formed by the vowels in the formant space [10, 17]. FCR and VAI are reciprocal of each other, with FCR describing how centralized the vowels are and VAI generally depicting the dispersion of vowels during speech [14]. AVS measures the size of the articulatory vowel space by calculating the average Euclidean distances, or the quantitative separation between two vowels in the formant space, for all possible vowel pairs [17]. FnRR encompasses a list of formant measures that describe the tongue mobility within the oral cavity during vowel production, as well as jaw lowering and lip movement. For example, F1RR and F2RR generally reflect the superior-inferior and anterior-posterior movement of the tongue dimensions inside oral cavity, respectively, during production of different vowels [14, 16, 18].

These derived formant metrics have been extensively used to study articulation associated in various pathological populations. These included patients with amyotrophic lateral sclerosis (ALS) [18], post-stroke dysarthria [14, 19, 20], hearing impairment [21], and post-lingually deaf speakers with cochlear implant [17], individuals without larynx [22], children with epiglottic cysts [23], as well as children with cerebral palsy [24]. In other studies, the metrics were used to correlate swallowing function in stroke patients with dysarthria [25], to distinguish speakers of different sexes and ages [26], and to distinguish between children with autism spectrum disorders from Down syndrome with intellectual disability [27]. Liu and colleagues [28] even developed ways to automatically calculate these formant metrics.

Despite many studies on PD vowels, previous findings remain inconclusive. The majority of these studies involved only small groups of PD participants with mild to moderate hypokinetic dysarthria (cf. [29–31]), limiting the generalizability of the findings to the broader PD population. Moreover, existing studies only examined the relationship between prosody distortion and PD progression, and distortion of speech was not fully investigated [9, 32]. Most of these studies were based on English-speaking PD population, leaving a knowledge gap concerning Chinese PD patients.

Given the significant language differences between English and Chinese, further research involving tonal languages such as Chinese is necessary. For a tone language such as Cantonese, the same phonetic segment will carry different meanings if produced at different lexical tones. In addition, Chinese is a syllable-timed language while English is a stress-timed language. Chinese tends to exhibit smaller variation in syllable duration and structure, realizations of stress, fewer stress-related rules, and more unstressed syllables as compared to stress-timed languages. To address the knowledge gap, the present study examined vowels produced by Cantonese PD speakers of different severity levels by making use of an array of derived formant metrics. Findings would provide a comprehensive understanding of how hypokinetic dysarthria associated with PD affects vowel production among Chinese PD patients throughout disease progression. In particular, more information can be gathered with regard to the tonal aspect of the language. This information can then help develop better assessment and intervention strategies for PD.

Methods

Participants

Eighteen adults (13 male and 5 female) aged between 56 and 83 years (mean = 64.61 years, SD = 8.25 years) who were diagnosed with idiopathic PD were recruited from the Hong Kong Parkinson’s Disease Association, which is a local PD patient group in Hong Kong. Among them, 11 were rated with mild PD (stage I–II) (2–14 years post onset) and 7 moderate PD (stage III–IV) (4–14 years post onset) based on the Hoehn and Yahr Scale [33]. Another group of 30 healthy individuals (HC) (15 M and 15 F) aged between 45 and 83 years (mean = 60.77 years, SD = 12.97 years) were also recruited to serve as controls in the study. The unbalanced pools of PD and healthy participants were mainly due to the difficulty in identifying suitable PD participants, especially those of greater PD severity as they tended to be immobile and wheel-chair-bound. All participants were native speakers of Cantonese and were physically healthy with no history of speech and hearing impairment, except for PD participants. Both PD and HC groups were age and gender matched. This human study was approved by Faculty Research Ethics Committee, Faculty of Education, University of Hong Kong – Approval No. 12/2017. All adult participants provided written informed consent to participate in this study.

Materials

Speech samples included the three corner vowels of Cantonese (/i/, /u/, and /a/) produced in the /kV/ syllable. The syllable structure /kV/ is selected as there are real words associated with /ki/, /ku/ and /ka/. To maintain naturalness of production, the syllable was embedded in the carrier phrase /ŋɔ jiu tʊk ___ peɪ neɪ tʰɛŋ/ (meaning “I need to read ___ to you”). The carrier phrase was selected so that the target vowel was preceded and followed by a stop, allowing easier identification of the vocalic segment during acoustic analysis. To avoid possible influence from lexical tones, all /kV/ syllables were produced at the high-level tone.

Instrumentation and Procedure

The recording procedures were identical for both PD and HC participants, with the exception that all speech samples from PD participants were recorded in the morning, about 60–90 min after the PD participants took their medication. This ensured that their performance was captured during the ON-phase of medication cycle. Each vowel, embedded in the carrier phrase, was produced three times, with the order of vowel production randomized.

During the recording, the participant was presented with cards on which the sentences (/kV/syllables embedded in the carrier phrase) were printed. The participants were instructed to read aloud the sentences at a comfortable pitch and loudness level to ensure naturalness of production. Direct modeling was provided as needed. A brief practice period was provided to each participant in order to familiarize themselves with the recording environment and materials. The recording took place in a quiet room of the Hong Kong Parkinson’s Disease Association. Speech samples were recorded using a condenser microphone (SM58, Shure, USA) which was placed at about 5 cm from the participant’s mouth. The microphone was connected via a preamplification unit (M-Track, M-Audio, USA) to a laptop computer. All audio signals were digitized at 44.1 kHz and 16 bits/sample and stored in a computer for later analyses using Praat (Version 6.0.25) [34].

Data and Statistical Analyses

To minimize the potential influences from preceding and following phonemes, only the medial 80% of the vowel was used to obtain F1 and F2 values. For each vowel, the F1 and F2 values were averaged before being used to calculate the derived formant metrics including tVSA (in Hz²), VAI, FCR, AVS (in Hz), F1RR, and F2RR. The formulas for calculating the six formant metrics can be found in Table 1.

Table 1.

Formulas of the derived formant metrics used in the study

Metrics	Formula
tVSA (in Hz2)	$\frac{1}{2}·\left|{F1}_{/i/}·\left({F2}_{/a/}-{F2}_{/u/}\right)+{F1}_{/a/}·\left({F2}_{/u/}-{F2}_{/i/}\right)+{F1}_{/u/}·\left({F2}_{/i/}-{F2}_{/a/}\right)\right|$
VAI	$\frac{{F2}_{/i/}+{F1}_{/a/}}{{F2}_{/u/}+{F2}_{/a/}+{F1}_{/i/}+{F1}_{/u/}}$
FCR	$\frac{1}{\text{VAI}}$
AVS (in Hz)	$\frac{2}{n\left(n-1\right)}·{\displaystyle \sum }_{i=1}^{n-1}{\displaystyle \sum }_{j=i+1}^n\sqrt{{\left({F1}_i-{F1}_j\right)}^2+{\left({F2}_i-{F2}_j\right)}^2}$
F1RR	$\frac{{F1}_{/a/}}{\frac{1}{2}·\left({F1}_{/i/}+{F1}_{/u/}\right)}$
F2RR	$\frac{{F2}_{/i/}}{{F2}_{/u/}}$

F1, first formant frequency; F2, second formant frequency.

In the present study, two-way analyses of variance were carried out for each derived formant metric to identify possible differences among different speaker groups (mild PD, moderate PD, and HC), and genders (males and females), and their interaction effect. Subsequent post hoc pairwise comparisons were carried out when necessary. All statistical procedures adopted a pre-set significance level of p = 0.05. Due to the known gender differences, no attempt was made to combine male and female data in the analyses.

Results

The average and standard deviation values of F1 and F2 associated with the three corner vowels produced by male and female HC, mild PD and moderate PD speakers are shown in Table 2. Based on the formant frequencies, the vowel space formed by the three corner vowels associated with male and female speakers are plotted in Figures 1 and 2, respectively. According to the figures, comparing the vowel spaces of healthy speakers and speakers with mild and moderate PD, PD speakers exhibited a smaller vowel space, and the difference appeared to be more as PD severity increased, as indicated by the smaller vowel space for moderate PD. The diminished vowel space indicated an undershooting for vowel articulation in PD speakers when compared with the healthy controls. In addition, the changes in vowel space size with PD seem consistent between males and females.

Table 2.

F1 and F2 values (in Hz) associated with the three corner vowels /i, a, u/ produced by healthy speakers and speakers of mild and moderate PD

​	Average (and standard deviation) formant frequencies (in Hz)
	male			female
	healthy controls	mild PD	moderate PD	healthy controls	mild PD	moderate PD
/i/
F1	300.76 (32.69)	350.48 (49.29)	338.04 (23.63)	338.51 (44.31)	360.45 (28.41)	371.68 (41.93)
F2	1,645.89 (277.60)	1,722.17 (126.40)	1,279.20 (541.03)	1,697.55 (495.77)	1,389.55 (102.92)	1,447.06 (525.39)
/u/
F1	360.73 (31.50)	400.40 (42.16)	379.06 (13.07)	393.08 (41.16)	415.61 (68.05)	415.38 (29.02)
F2	707.53 (69.95)	820.55 (90.75)	775.26 (133.98)	773.24 (76.39)	864.81 (105.07)	839.37 (13.24)
/a/
F1	745.90 (62.49)	695.78 (93.36)	586.97 (116.05)	836.93 (98.26)	894.39 (225.14)	704.24 (105.85)
F2	1,211.19 (91.63)	1,159.17 (100.67)	1,079.25 (80.71)	1,270.14 (153.59)	1,286.25 (99.86)	1,118.20 (146.60)

Fig. 1.

Vowel space associated with the three corner vowels /i, a, u/ produced by male healthy speakers, and speakers with mild and moderate PD.

Fig. 2.

Vowel space associated with the three corner vowels /i, a, u/ produced by female healthy speakers, and speakers with mild and moderate PD.

Based on the F1 and F2 values obtained, the six derived formant metrics were calculated, and the results are shown in Table 3. For those metrics describing size of vowel space including tVSA, VAI, and AVS, and those for lip and jaw movement such as F1RR and F2RR, they consistently demonstrated a trend: healthy speakers exhibited the highest values, followed by speakers with mild PD, and speakers with moderate PD showed the smallest values, and the trend is present in both male and female speakers. This is consistent with what is observed from the vowel spaces shown in Figures 1 and 2. For FCR, as it is the reciprocal of VAI, it tended to increase with severity of PD.

Table 3.

Derived formant metrics associated with healthy speakers, and speakers with mild and moderate PD

​	Average and standard deviation values of derived formant metrics
	male			female
	healthy controls	mild PD	moderate PD	healthy controls	mild PD	moderate PD
tVSA (in Hz2)	194,941.31 (56,224.78)	140,767.40 (42,414.09)	76,517.23 (64,925.83)	216,547.48 (120,292.36)	147,440.43 (105,009.88)	104,952.28 (104,549.63)
VAI	0.93 (0.12)	0.89 (0.08)	0.72 (0.24)	0.91 (0.16)	0.78 (0.09)	0.78 (0.19)
FCR	1.09 (0.15)	1.13 (0.11)	1.51 (0.53)	1.13 (0.25)	1.29 (0.15)	1.34 (0.37)
AVS (in Hz)	748.01 (150.47)	677.77 (83.13)	494.29 (246.45)	792.24 (253.92)	570.37 (219.11)	552.11 (239.91)
F1RR	2.27 (0.25)	1.86 (0.23)	1.64 (0.33)	2.31 (0.33)	2.29 (0.30)	1.78 (0.31)
F2RR	2.35 (0.46)	2.12 (0.24)	1.64 (0.59)	2.23 (0.72)	1.63 (0.32)	1.72 (0.61)

With regard to inferential statistics, analyses of variances revealed no significant interaction between speaker group and gender for all derived formant metrics (ps > 0.1). However, a significant main effect was found for speaker group for all formant metrics, tVSA (F(2, 42) = 5.67, p = 0.006, ηp2 = 0.46), VAI (F(2, 42) = 4.25, p = 0.021, ηp2 = 0.41), FCR (F(2, 42) = 4.60, p = 0.011, ηp2 = 0.43), AVS (F(2, 42) = 5.13, p = 0.010, ηp2 = 0.44), F1RR (F(2, 42) = 11.77, p < 0.001, ηp2 = 0.60), and F2RR (F(2, 42) = 4.27, p = 0.021, ηp2 = 0.41), suggesting at least one speaker group was significantly different from the others for all formant metrics. Subsequent post hoc pairwise comparisons revealed that, for all formant metrics, HC speakers were significantly different from PD speakers: F1RR (p < 0.001), tVSA, AVS (p < 0.005), F2RR (p < 0.01), VAI, and FCR (p < 0.05). Vowel characteristics between HC and moderate PD speakers were also significantly different for all metrics: F1RR, tVSA, AVS, FCR, VAI, and F2RR (all ps < 0.05). F1RR and tVSA associated with mild PD were significantly lower than HC (ps < 0.05), while F1RR of mild PD was significantly greater than moderate PD speakers (p < 0.05). For all formant metrics, no significant main effects were found for gender, indicating comparable vowel distribution between male and female speakers.

Discussion

The presence of hypokinetic dysarthria in PD patients, which greatly affects their speech production, has been widely documented. It should be noted that the ability of PD speakers to accurately articulate speech sounds is compromised, greatly reducing their speech intelligibility. The present study examined vowels produced by Cantonese PD speakers using derived formant metrics, aiming to uncover more subtle changes in their ability to precisely produce Cantonese vowels compared with healthy controls. The use of these formant metrics was to acoustically reveal the distribution of vowels in the vowel space, in correlation with presence and severity of PD. Centralization, dispersion, and articulation of Cantonese vowels produced by PD speakers along their disease progression were thus unveiled.

Articulation of Vowels

The findings of the diminished F1 and F2 confirmed the presence of articulatory undershooting and a reduction in working vowel space in PD speakers when compared with HC, and PD severity appeared to correlate with the extent of reduction. Regardless of gender, PD speakers generally exhibited a reduction in tongue height and tongue advancement during vowel production, as revealed by the elevation in F1 and F2 values when compared with healthy speakers. This finding was further supported by the reduced F1RR and F2RR values, implying a reduction in jaw opening and lip movement. The finding of a reduced F2RR in PD speakers is in line with Sapir et al. [2, 14]. The reduced jaw opening and lip movement during vowel production appeared to be more apparent in speakers of moderate PD. F1RR describes alteration to the tongue height dimension [18], yet it has been used infrequently in related literature. Fougeron and Audibert [18] examined the F1RR values associated with vowels produced by French dysarthric patients, and they found that F1RR was significantly reduced in male speakers with flaccid-spastic dysarthria due to ALS. Findings from the present study might add to our understanding of the use of F1RR in describing vowel articulation.

Working Vowel Space

Recall that the literature presents inconsistencies regarding tVSA in relation to PD, with some researchers being skeptical of its sensitivity in describing articulation among PD patients (cf. [12, 16]). However, the results from the present study revealed a marked reduction in tVSA associated with mild and moderate PD when compared with HC speakers. This is consistent with the reduced jaw lowering and lip movement during vowel production. In addition to the compressed working vowel space, dispersion of PD vowels as indicated by AVS was reduced when compared with healthy controls. This contradicts the vowel variability study of PD speakers by de Paula Soares [29]. However, it should be noted that there were only six (3 male and 3 female) PD patients in that study, and PD severity was not mentioned, resulting in a lack of representativeness of her findings. In addition, the PD individuals spoke Dutch and Brazilian Portuguese, which was different from Cantonese used in the present study. Future studies should examine this aspect more systematically by recruiting more PD speakers of different severities.

Vowel Centralization

The reduced working vowel space was further supported by the increased FCR and reduced VAI. FCR serves to reflect how centralized the vowels are within the formant space. The greater FCR associated with PD speakers implied that, in addition to the reduced working vowel space; the vowels were more packed in the vowel space. This indicated a restricted articulatory space in PD speakers, when compared with healthy speakers. As VAI is simply the reciprocal of FCR, they should be inversely proportional to each other. Whereas FCR increases, VAI decreases. Existing literature indicates that both VAI and FCR were effective in differentiating working vowel spaces between PD speakers with hypokinetic dysarthria and healthy controls [9, 14]. In fact, Skodda and colleagues [9] suggested that VAI was effective in detecting and identifying the subclinical changes in vowel production even in mild PD patients, and VAI appeared to be a better measure than tVSA in describing the vowel impairment associated with PD, and VAI could serve as a sensitive biomarker for PD progression. Apparently, more studies, especially those involving PD patients of different severities, are needed in order to confirm this suggestion.

Due to the motoric problems associated with PD including muscle rigidity and bradykinesia, PD speakers experience slowness and inadequacy of motor movement, and that also affects the various aspects of speech production including hypokinetic dysarthria [35]. As a result, PD speakers experience varying difficulties in accurately articulating different vowels. The general inadequacy appeared to lie in their tongue and jaw movement, ending up in articulatory undershooting. This is confirmed by the formant metrics showing a diminished working vowel space. In addition, F1RR and F2RR values further indicated that PD speakers tended to exhibit restricted jaw opening and lip movement. The present study again confirmed the value of using these formant metrics in quantifying vowel articulation in PD.

Gender Effect on Formant Metrics

The present findings did not find any gender effect for all formant metrics. Despite the observed differences in F1 and F2 values between men and women [36, 37], the derived formant metrics which were used to describe the distribution of vowels and differences in articulation were not gender sensitive. This appears to indicate the robustness of these formant metrics, allowing to reveal the net effect of presence and severity of PD to vowel articulation.

Clinical Implications and Future Research

While the present study focused on acoustic changes in vowel production, these changes have direct and meaningful implications for speech intelligibility, which is a primary clinical concern in PD. The observed reduction in the working vowel space (evidenced by decreased tVSA and AVS values), increased vowel centralization (indicated by elevated FCR), and restricted articulatory movement (reflected in lower F1RR and F2RR) collectively contribute to the perceptual phenomenon of vowel blurring.

Acoustically, a compressed and centralized vowel space reduces the perceptual distinctiveness between different vowel categories. For example, a reduced F1RR indicates less jaw lowering for the open vowel /a/, causing it to sound closer to mid or high vowels. Similarly, a reduced F2RR suggests less extreme tongue fronting/backing, diminishing the contrast between vowels like /i/ and /u/. This loss of acoustic contrast is a known contributor to reduced intelligibility, as listeners rely on clear vowel targets to decode speech signals, especially in noisy environments or without visual cues.

It follows that tVSA could be used as visual biofeedback during speech therapy. In practice, software could provide patients with a real-time display of their vowel space during exercises targeting corner vowels (/i, a, u/). The goal of “expanding the triangle” on screen offers concrete, objective reinforcement for increasing articulatory range, complementing traditional auditory-perceptual feedback. This could be integrated into established therapies like LSVT LOUD^® to help differentiate gains in loudness from gains in articulation.

Furthermore, serial tVSA measurements offer an objective method for tracking therapeutic progress and quantifying change over time. For wider implementation, developing streamlined clinical software and establishing language-specific normative references will be essential. By linking these sensitive acoustic metrics to functional communication outcomes, this research supports the development of more targeted, measurable, and effective speech interventions for individuals with PD.

Previous research has established links between vowel space metrics and intelligibility ratings in dysarthria. Studies on hypokinetic dysarthria in PD have shown that smaller vowel space areas correlate with poorer listener intelligibility scores. The sensitivity of tVSA and F1RR in detecting even subtle changes in mild PD, as found in our study, suggests these metrics could serve as early markers of declining intelligibility, potentially before it becomes subjectively apparent to patients or clinicians.

From an intervention perspective, these findings support the rationale for therapies that target articulatory precision and range of motion, such as Lee Silverman Voice Treatment (LSVT^® LOUD), which emphasizes increased vocal effort and amplitude, thereby indirectly expanding vowel space. Our results suggest that tVSA could be a sensitive, objective biofeedback tool in such therapy, allowing patients and clinicians to monitor improvements in articulatory working space alongside perceptual intelligibility gains.

Future research should directly correlate these derived formant metrics with perceptual intelligibility scores in Cantonese-speaking PD patients to establish language-specific acoustic-perceptual relationships. This would further solidify the clinical utility of these measures in assessment and treatment monitoring.

Conclusion

The present study attempted to quantify the change in vowel distribution associated with mild and moderate PD speakers of Cantonese. Findings suggest a compressed working vowel space in PD, and the extent appeared to correlate with disease progression. Specifically, as revealed by the formant metrics, PD speakers exhibited a diminished vowel space contributed by the restricted jaw opening and tongue movement during vowel production, which was confirmed by other formant metrics. From another perspective, PD speakers demonstrated a reduced dispersion in vowels. Such change in vowel articulation appeared greater for moderate than mild PD, and it understandably contributed to the reduced intelligibility of PD speech. It is suggested that tVSA and F1RR are sensitive formant metrics for describing the subtle changes in vowel articulation in PD. And tVSA may serve as faithful biofeedback in treatment for mild PD speakers for improving patients’ speech intelligibility.

Statement of Ethics

This human study was approved by Faculty Research Ethics Committee, Faculty of Education, University of Hong Kong – approval: 12/2017. All adult participants provided written informed consent to participate in this study.

Conflict of Interest Statement

The authors have no conflicts of interest to declare.

Funding Sources

The study was funded by the Education Faculty Research Fund of the Faculty of Education, University of Hong Kong (Ref No. 000250337).

Author Contributions

Shi Zhu and Yang Chen: substantial contributions to the conception, design of the study, and manuscript preparation. Natalie Leung and Eric T.-S. Tong: substantial contributions to the subject recruitment, data collection and analysis, statistical analyses, and manuscript preparation. Manwa L. Ng: substantial contributions to the conception and design of the study, statistical analyses and manuscript preparation.

Funding Statement

The study was funded by the Education Faculty Research Fund of the Faculty of Education, University of Hong Kong (Ref No. 000250337).

Data Availability Statement

All data generated or analyzed during this study are included in this article. Further inquiries can be directed to the corresponding author.