Description of PD Phonation in Terms of EEG-Related Frequency

Bands

Pedro Gómez

, Jiří Mekyska

, Luboš Brabenec

, Patrik Šimko

Irena Rektorová

Andrés Gómez

and Victoria Rodellar

NeuSpeLab, CTB, Universidad Politécnica de Madrid, 28220 Pozuelo de Alarcón, Madrid, Spain

Department of Telecommunications, Brno University of Technology, Brno, Czech Republic

Applied Neuroscience Research Group, Central European Institute of Technology - CEITEC,

Masaryk University, Brno, Czech Republic

First Department of Neurology, Faculty of Medicine and St. Anne’s University Hospital,

Masaryk University, Brno, Czech Republic

Usher Institute, Faculty of Medicine, University of Edinburgh, Edinburgh, U.K.

{lubos.brabenec, patrik.simko, irena.rektorova}@ceitec.muni.cz, a.gomezrodellar@ed.ac.uk

Keywords: Parkinson’s Disease, Functional Assessment of Phonation, Neuromotor EEG Activity Monitoring, Repetitive

Transcranial Magnetic Stimulation.

Abstract: Parkinson’s Disease (PD) is an increasing prevalence neurodegenerative condition affecting the life quality

of people suffering from its neuromotor and cognitive performance. PD symptoms include vocalization and

speech alterations, known as hypokinetic dysarthria (HD). One of the manifestations of HD is unstable

phonation. Repetitive Transcranial Magnetic Stimulation (rTMS) is a non-invasive method that may improve

some motor and non-motor symptoms of persons with PD (PwP). The present study concentrates on analyzing

and comparing the phonation behavior of two cases before (pre-stimulus) and after (post-stimulus) ten

sessions of rTMS treatment, to assess the extent of changes in their vocalization. Voice recordings of a

sustained vowel [a:] taken immediately before and after the treatment, and at follow-up sessions (at six, ten,

and fourteen weeks after the baseline assessment) were processed by inverse filtering to estimate a

biomechanical correlate of vocal fold stiffness, which band-pass filtered into EEG-related frequency bands.

Log-likelihood ratios between pre- and post-stimulus amplitude distributions of each frequency band, Mann-

Whitney U-tests, and normalized difference scores showed significant improvements in the actively

stimulated case, which were not observed in the sham case. Early preliminary insights into the capability of

phonation quality assessment on monitoring neuromechanical activity from acoustic signals are shown.

1 INTRODUCTION

Parkinson’s Disease (PD) is a neurodegenerative

disorder with a prevalence of around 200 cases per

100,000 persons, at a growing incidence rate of 15

cases per 100,000 (Dorsey et al., 2007). It has a severe

impact on the life quality of persons with PD (PwP)

https://orcid.org/0000-0003-3283-378X

https://orcid.org/0000-0002-6195-193X

https://orcid.org/0000-0002-8348-5757

https://orcid.org/0000-0002-0767-0918

https://orcid.org/0000-0002-5455-4573

https://orcid.org/0000-0003-3283-378X

https://orcid.org/0000-0001-9384-3290

affecting motor and non-motor symptoms (Duffy

2013). Perturbed respiration, phonation, articulation,

and prosody are among motor symptoms hampering

vocalization and speech, in what is known as

hypokinetic dysarthria (HD), characterized by mono-

pitch and mono loudness, imprecise articulation,

impaired speech rate, and rhythm, and irregular pitch

ﬂuctuations. Repetitive transcranial magnetic

226

Gómez, P., Mekyska, J., Brabenec, L., Šimko, P., Rektorová, I., Gómez, A. and Rodellar, V.

Description of PD Phonation in Terms of EEG-Related Frequency Bands.

DOI: 10.5220/0011669100003414

In Proceedings of the 16th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2023) - Volume 4: BIOSIGNALS, pages 226-233

ISBN: 978-989-758-631-6; ISSN: 2184-4305

 2023 by SCITEPRESS – Science and Technology Publications, Lda. Under CC license (CC BY-NC-ND 4.0)

stimulation (rTMS) is a non-invasive method used to

modulate neuronal excitability which has been

proposed as a therapy to improve various symptoms

of PD (Brabenec et al. 2019). The purpose of the

present study is two-fold: on the one hand, to explore

the relationship between neuromotor activity and

phonation through the inversion of a cortico-muscular

coupling (CMC) model (Branbilla 2021); on the other

hand, to show how to characterize and monitor the

efficiency of rTMS in two study cases using CMC

model inversion.

The structure of the paper is as follows. Section 2

is devoted to explaining the conducting narrative

capitalizing on the possibility of describing

phonation-related biomechanical correlates as vocal

fold stiffness in terms of EEG-related frequency

bands based on the well-known relationship between

EEG neuroelectrical activity in the premotor and

supplementary motor cortex and the neuromuscular

activity in the laryngeal nerves controlling voice

production explained by CMC (McKeown et al,

2006). Section 3 describes the signal inversion

methods to determine neuromotor activity in EEG-

related frequency bands, and their statistical

distributions in two cases of PwP submitted to active

and sham rTMS. Section 4 compares the results from

the two study cases, which are discussed in Section 5.

Contributions, findings, and conclusions are

disclosed in Section 6.

2 FUNDAMENTALS

The objective of the present study is to dive into the

relationship between neuromotor and acoustical

activity (neuroacoustical) involved in vocalization

(phonation and articulation) with application to the

characterization of PD hypokinetic dysarthria, as

summarized in 0. The simplified neuroacoustical

model (acoustical and neuroneuromotor) of the

system controlling vocalization (top-down view) is

summarized in 0.a. The vocalization structure

involves the lungs, larynx, and the oro-naso-

pharyngeal cavities, considering the lips as the

radiation place where the speech wave is projected to

the surrounding media. Important muscular systems

control each substructure, such as the diaphragm and

intercostal (not shown) controlling lung pressure, the

laryngeal muscles (thyroarytenoid, cricothyroid, and

transverse and oblique arytenoid) of which the

thyroarytenoid (musculus vocalis) is responsible for

blocking and releasing airflow through the glottis and

producing the basic vibration in voiced speech

(phonation), the hypoglossus, extrinsic and intrinsic

glossal, controlling tongue movements, the

mandibular, regulating jaw raising and lowering, and

the orofacial, defining lip rounding gestures.

Figure 1: Acoustical and neuromotor pathways in

vocalization: a) Top-down model from the neuromotor to

the acoustical (speech wave); b) Bottom-up model from

phonation to neuromechanical activity estimation.

The neuromotor activity (NMA) controlling each

group of muscles is driven by common areas of

activation for semantic processing, lexical selection,

syntactic construction as well as oral articulation,

involving mainly the periaqueductal, premotor, and

supplementary motor brain areas, among others

(Schulz et al., 2005). These areas project to the

specific muscles by different neural pathways, such

as the phrenic nerve (diaphragm), the superior and

recurrent laryngeal (laryngeal muscles), the

pharyngeal, hypoglossus, extrinsic, and intrinsic

tongue nerves (oropharyngeal), and the trigeminal

branch controlling mandibular and orofacial muscles.

The inverse bottom-up model shown in 0.b provides

a view of the model inversion methodology proposed

in the present study to estimate and visualize NMA

involved in phonation. The speech signal is processed

by a linear predictive algorithm (Deller, Proakis, and

Hansen, 1993) to produce a glottal source correlate.

The inverse filter used in the processing is used to

estimate acoustic-articulatory features, such as

formants, vowel triangle aspect ratios, tongue-jaw

kinematic movement, masseter NMA, etc. (Gómez et

al., 2021). Among the mentioned larynx muscles, the

thyroarytenoid is of special interest for the study

because it is a very small mass low-inertial muscle,

responding well to vibration frequencies beyond

Description of PD Phonation in Terms of EEG-Related Frequency Bands

227

1kHz, and is directly related to phonation. The study

capitalizes on the estimation of the thyroarytenoid

muscle stiffness on the glottal source correlate, which

is used to produce estimations of the vocal fold

biomechanical features. It is assumed the vocal fold

stiffness to be controlled by the inferior laryngeal

nerve NMA on the thyroarytenoid muscle, therefore,

it may be used to estimate EEG-related frequency

bands in the laryngeal nerve periaqueductal control.

This inverse bottom-up chain could help in better

monitoring phonation control in PwP (Rektorova et

al., 2012). Eventual instabilities of the laryngeal

NMA will appear as oscillations in the frequency

bands (tremors).

3 MATERIALS AND METHODS

The present research includes results from two PwP

participants (under active and sham rTMS) in a study

devised to HD in PwPs (Brabenec et al., 2021). The

NMA of both participants was monitored at several

evaluation periods since stimulation using sustained

emissions of the vowel [a:], and the results have been

compared.

Table 1: Participants’ demographic and clinical data; A:

active stimulation; S: sham stimulation; M: Male Gender;

Y: years. UPDRS-III: Unified Parkinson’s Disease Rating

Scale part III (motor).

PwP code

(pre)

Active/Sham Gender Age (Y) UPDRS

-III

1400 A M 64 10

1900 S M 77 8

The speech processing was based on fragments of

4s long vowel emissions between the time instants at

2s and 6s from the vowel onset, sampled at 16 kHz

and 16 bits. An inverse lattice-ladder filter (Deller,

Proakis, and Hansen, 1993, Alku et al., 2019)

evaluated the oropharyngeal tract model, estimating

the residual prediction error, which once integrated,

produced the glottal source correlate. The vocal fold

stiffness (VFS) was estimated from the glottal source

correlate adjusting its spectral power by a 2-mass

model of the vocal fold biomechanics (Gómez et al.

2009) The VFS was de-biased and de-trended by a

moving-average filter. See 0 for an example of

unbiased VFS (UVFS) estimation from case 1400.

The UVFS was band-pass filtered at the EEG-related

frequency bands (δ: f≤4 Hz; ϑ: 4 Hz<f≤8 Hz; α: 8

Hz≤f≤16 Hz; β: 16 Hz<f≤32 Hz; γ: f>32 Hz; μ: 8

Hz<f≤12 Hz), producing a set of UVFS estimates

given by 𝜉





(𝑛), where i=(0,..., I) is the session index

(I=5), j=a (active case) or j=s (sham case) is the

participant index, and k=(1 for δ,…, 6 for μ) is the

frequency band index of the six frequency bands

defined above, and n is the time index. 0 shows the

results of the band-pass separation in the time domain

(left column) and their power spectrograms (right

column). The distributions p{ξ} of EEG-related

frequency bands were estimated from normalized

amplitude histograms as

𝑝

,



(

𝜉

)

=ℎ𝜉

,



(

𝑛

)



(1)

The distributions from post-stimulus recordings were

compared on their overlap interval ξ ϵ Ω with pre-

stimulus ones using log-likelihood ratios

𝜆

,



(

𝑝



|𝑝



)

= 𝑙𝑜𝑔𝑝

,



(

𝜉

)

𝑝

,



(

𝜉

)

𝑑𝜉



(2)

as well as Mann-Whitney U-tests (T

)

𝑇

,



=𝑇



𝜉

,



(

𝑛

)

,𝜉

,



(

𝑛

)



(3)

and time-weighted scores (𝑠





and 𝑠





)

𝑠

,



〈

𝜉

,



(

𝑛

)〉

−

〈

𝜉

,



(

𝑛

)〉

〈

𝜉

,



(

𝑛

)〉

𝑤



;

𝑤



=(𝑑



−𝑑



)(𝑑



−𝑑



)

⁄

(4)

where the weight w

is a normalizing factor to take

into account the time interval between each post-

stimulus date d

and the corresponding pre-stimulus

date d

normalized to the longest interval (d

- d

) in

days, therefore, long-lasting beneficial effects were

given larger importance than short-duration effects.

The time intervals in days for the case 1400 were

T1=15, T2=47, T3=74, and T4=99. The

corresponding ones for case 1900 were T1=9, T2=42,

T3=65, and T4=96. Finally, the VFS unbalance (the

difference between two neighbor estimates of the

total UVFS relative to their average) is added as a

reference feature indirectly related to the jitter

𝑢

,

𝜉

,

(

𝑛

)

−𝜉

,

(

𝑛−1

)

𝜉





(

𝑛

)

+𝜉





(

𝑛−1

)

(5)

4 RESULTS

Case 1400 corresponded with a participant who was

submitted to active stimulation. The pre-stimulation

recording of a sustained vowel [a:] produced an

BIOSIGNALS 2023 - 16th International Conference on Bio-inspired Systems and Signal Processing

228

Figure 2: Example of the vocal fold body stiffness estimation from a segment of 4 s of phonation from case 1400, during the

utterance of a sustained vowel [a:]: a) speech segment showing two events of phonation blocking at intervals 4.0 s - 4.2 s and

5.6 s - 5.8 s, where the f0 drops down from 120 Hz to 60 Hz; b) vocal fold stiffness (black), its detrend tendency (red) and its

unbias (blue).

utterance with a duration of 11.163 s, during which

five events of phonation blocking were observed. The

segment selected for the analysis was extended

between 3 s and 7 s, to include the first two blocking

events, which are shown in 0.a, with an estimation of

its fundamental frequency (f0) profile superimposed

in red. 0.b shows the estimation of the vocal fold

stiffness (black), its detrend (red), and the unbiased

detrended result (blue). In its turn, 0 depicts the

results of splitting the unbiased vocal fold stiffness

(UVFS) into the EEG-related frequency bands

defined in Section 3 in the time domain (left column)

and as spectrograms (right column), the first one (0.b)

being given in a logarithmic representation to

compensate different amplitude levels, and the

remnant ones are shown in linear representation. The

distributions of the pre-stimulus and post-stimulus

EEG-related frequency bands are presented in 0.In

Figure 2 an example of VFS estimation from the

acoustic signal is shown, where two blocking events

are appreciated as 4 s and 5.2 s. These events are

marked by a descent in f0 (red line in Figure 2.a) from

120 Hz to 60 Hz during approximately 200 ms, during

which, the estimation of f0 becomes erratic and

unstable. This behavior is aligned with the UVFS

profile shown in Figure 2.b, where f0 instability is

explained by a decay in the VFS at 4 s, followed by a

correction action at 4.3 s to initiate a new slow decay

to be corrected again at 5.6 s.

The graphical example showing the frequency band

estimation protocol illustrated in Figure 3 adds new

information to the evolution of these events. The

activity on the δ-band in Figure 3.c shows the decay

starting really at 3.7 s, coming to a minimum at 4.1 s,

and being strongly incremented immediately after,

followed by a slow progressive decay and correction

bursts between 5.0 s and 5.6 s, where a new pull-up is

observed. The corrective actions are evident in all the

bands, but the most interesting ones are β and γ, given

in Figure 3.j and l. The β-band shows a strong activity

burst between 22-30 Hz at 4.1-4.3 s, followed by a

narrower one at 26 Hz and 4.4-4.7 s. The γ-band, on

the contrary, shows a narrow burst at 38 Hz between

4.4-4.5 s, and a wider and stronger one at 42-46 Hz

lasting from 5.4-5.7 s. According to recent research,

the activity in these bands could be related to the

neuromotor correction (β), and interaction of

different brain areas (γ), including the striato-cortical,

the cerebellum, and the temporal auditory areas

(Ibarra-Lecue, Haegens, and Harris, 2022).

The information provided in Figure 4.a and b,

corresponding to the activity on the β-band is also

quite meaningful, as it shows the profile of the

distributions for the five recordings corresponding to

the active case 1400. Whereas the first evaluation

(1400, pre-stimulus) is amplitude-widespread, all

post-stimulus evaluations (T1-T4) concentrate in

small amplitudes, all of them showing a χ2 profile.

The averages and standard deviations of the post-

stimulus evaluations are sensibly smaller than the pre-

stimulus one, pointing to a strong instability reduction

after stimulation. By contrast, the five recordings

corresponding to the sham case 1900 given in Figure

4.c and d, do not show any meaningful changes,

except possibly in the second evaluation (T1, 1900).

Description of PD Phonation in Terms of EEG-Related Frequency Bands

229

Figure 3: Spectral contents of the EEG-band description of 4 s of phonation from case 1400: a) unbiased vocal fold stiffness;

b) spectrogram of the vocal fold stiffness; c) δ-band component; d) spectrogram of the δ-band; e-f) id. of the ϑ-band; g-h); id.

of the α-band; i-j) id. of the β-band; k-l) id. of the γ-band; m-n) id. of the μ-band.

Figure 4: Probability density distributions of the EEG β-band activity from cases 1400 and 1900: a) pdfs from each evaluation

of the active case (1400: T0-T4); b) their corresponding boxplots; c-d) id. from the sham case (1900: T0-T4).

BIOSIGNALS 2023 - 16th International Conference on Bio-inspired Systems and Signal Processing

230

The results of the comparisons between pre-

stimulus and post-stimulus frequency distributions

estimated following expressions (2)-(3) are given

inTable 2 and Table 3. The results of evaluating the

improvement scores following expression (4) are

given in 0. 0 summarizes graphically the amplitude

averages of the five non-overlapping frequency bands

(δ, ϑ, α, β, and γ) in the five evaluation instants (T0-

T4), and the VFS unbalance (as a reference) for the

two cases being considered, their relative differences

concerning T0, and the same differences weighted by

the in-between-evaluation time intervals, following

expression (4).

Table 2: LLR scores according to expression (2) on the β-

band of post-stimulus evaluations (T1-T4) relative to pre-

stimulus (T0).

PwP code T1 T2 T3 T4

1400 0.675 0.607 0.469 0.593

1900 0.192 -0.005 0.083 -0.015

Table 3: p-values on the β-band from MW U-tests of post-

stimulus evaluations (T1-T4) relative to pre-stimulus (T0)

according to expression (3).

PwP code T1 T2 T3 T4

1400 <0.001 <0.001 <0.001 <0.001

1900 <0.001 <0.001 <0.001 0.036

Table 4: Improvement scores according to expression (4)

summarizing the β-band of post-stimulus evaluations

(T1-T4) relative to pre-stimulus (T0).

PwP code T1 T2 T3 T4

1400 -0.135 -0.418 -0.615 -0.878

1900 -0.058 -0.074 -0.233 -0.105

5 DISCUSSION

The aim of this study was focused on exploring the

capability of EEG-related frequency bands to explain

the activity on the neuromotor pathways related to

phonation using acoustic signals analyzing sustained

Figure 5: Normalized average values of each EEG-band (δ-γ) compared with the vocal fold stiffness unbalance from an active

stimulation case (1400) and a sham case (1900): a-b) absolute values; c-d) relative post-stimulus values (T1-T4) compared

with the pre-stimulus evaluation (T0); e-f) relative results weighted accordingly to the time interval between the pre-stimulus

and each post-stimulus interval, as defined in expression (4).

Description of PD Phonation in Terms of EEG-Related Frequency Bands

231

vowel vocalizations from PwPs submitted to active

and sham rTMS.

The comparison among pre- and post-stimulus

estimations in terms of LLRs given in 0 confirms the

observations on the β-band, pointing to strong

improvements in the active case (λ>0), whereas the

sham case shows mixed behavior and moderate

improvements in T1 and T3 which might be due to

circumstantial or confounding factors. The p-values

which are shown in 0 avail the estimations given in 0

for a significance level of 0.05 on the null hypothesis

of equal medians.

After examining the global improvement scores

on all the non-overlapping frequency bands given in

0, it may be concluded that taking the time interval

between the pre-stimulus and each post-stimulus

evaluation into account, the progress in the process

induced by rTMS seems steady, at least for the

observation time intervals considered. These findings

may be better examined on the evolution templates

given in 0.a and b, where the normalized amplitude

average values of the frequency-band components are

given, as well as the VFS unbalance regarding

expression (5) which is added as a reference. The

improvements of the phonation instability conditions

for the active case 1400 are evident (0.a), whereas the

evaluations from the sham case (0.b) do not show a

clear tendency. When considering the difference

between the pre-stimulus and each post-stimulus

estimations by bands given in 0.c and d, the droppings

observed in the active case (1400) become more

evident when compared with the random behavior of

the sham case (1900). This comparison is even more

meaningful when comparing the same differences in

all frequency bands weighted by the time intervals

between each pre- and post-stimulus pair, as seen in

0.e and f. The monotonous descent observed in 0.e is

indicative of the almost-permanent improvements

observed in the active case during the period

considered, contrasting with the quasi-erratic

behavior of the sham case.

The character of this study is very specific,

exploratory, and limited to the observations from the

two cases considered, and further efforts would be

required to generalize its potential application on a

large database.

6 CONCLUSIONS

The present paper is intended to explore the

possibilities of predicting the interactions on the

EEG-related β-γ frequency bands of the NMA from

the phonation acoustical signal. Albeit the specificity

of the cases studied is a limit to the findings observed,

the methodology proposed to extract neuromotor

activity from acoustical information to characterize

PwP vocalization may provide new meaningful

insights into the neuromotor activity related to

phonation stability. The three scores used in the

assessment of potential improvement behavior of

PwP phonation after active rTMS are in full

agreement, and can be used alternatively or

combined. These facts may open new applications of

signal processing in the field of speech neuromotor

understanding, and neurodegenerative disease

monitoring.

ACKNOWLEDGEMENTS

This research received funding from European

Union’s Horizon 2020 research and innovation

program under the Marie Skłodowska-Curie grant

agreement no. 734718 (CoBeN), a grant from the

Czech Ministry of Health, 16-30805A, a grant from

EU – Next Generation EU (project no.

LX22NPO5107 (MEYS)), and grants TEC2016-

77791-C4-4-R (Ministry of Economic Affairs and

Competitiveness of Spain), and Teca-Park-

MonParLoc FGCSIC-CENIE 0348-CIE-6-E

(InterReg Programme). Andrés Gómez-Rodellar

holds a scholarship from the Medical Research

Council Doctoral Training Programme in the Usher’s

Institute (University of Edinburgh Medical School).

REFERENCES

Alku, P., et al. (2019). OPENGLOT-An open environment

for the evaluation of glottal inverse filtering, Speech

Communication 107 (2019) 38-47. https://doi.org/

10.1016/j.specom.2019.01.005.

Brabenec, L. et al. (2021) Non-invasive brain stimulation

for speech in Parkinson’s disease: A randomized

controlled trial, Brain Stimulation, 14, 571-578.

https://doi.org/10.1016/j.brs.2021.03.010.

Brambilla, C. et al., (2021). Combined Use of EMG and

EEG Techniques for Neuromotor Assessment in

Rehabilitative Applications: A Systematic Review,

Sensors, 21 7014. https://doi.org/10.3390/s21217014.

Deller, J. R., Proakis, J. G., and Hansen, J. H. L. (1993)

Discrete-Time Processing of Speech Signals,

NewYork, Macmillan.

Dorsey, E. R., et al (2007). Projected number of people with

Parkinson's disease in the most populous nations, 2005

through 2030, Neurology 68(5) 384-386.

https://doi.org/10.1212/01.wnl.0000247740.47667.03.

BIOSIGNALS 2023 - 16th International Conference on Bio-inspired Systems and Signal Processing

232

Duffy, J. R. (2013). Motor Speech Disorders: Substrates,

Differential Diagnosis, and Management, 3rd Ed.,

Elsevier.

Gómez, P. et al. (2009). Glottal Source biometrical

signature for voice pathology detection, Speech

Communication 51(9) 759-781. https://doi.org/10.10

16/j.specom.2008.09.005.

Gómez, A., et al. (2021). Acoustic to kinematic projection

in Parkinson’s disease dysarthria. Biomedical Signal

Processing and Control 66 Art. 102422. https://doi.org/

10.1016/j.bspc.2021.102422.

Ibarra-Lecue, I., Haegens, S., and Harris, A. Z. (2022)

Breaking Down a Rhythm: Dissecting the Mechanisms

Underlying Task-Related Neural Oscillations, Front.

Neural Circuits, 16 846905. https://doi: 10.3389/

fncir.2022.846905.

McKeown, M. J., et al. (2006). Cortical muscle coupling in

Parkinson’s disease (PD) bradykinesia. Parkinson’s

Disease and Related Disorders, P. Reiderer et al. (eds.),

Springer, Vienna (2006) 31-40. https://doi.org/

10.1007/978-3-211-45295-0_7.

Mekyska, J., et al. (2015) Robust and complex approach of

pathological speech signal analysis, Neurocomputing

167 94-111. https://doi.org/10.1016/j.neucom.2015.02.

085

Rektorova, I., et al. (2012). Functional neuroanatomy of

vocalization in patients with Parkinson’s disease, J.

Neu. Sci. 313 7-12. https://doi.org/ 10.1093/cercor/bhi

06.

Schulz, G. M. et al. (2005). Functional neuroanatomy of

human vocalization: an H215O PET study, Cereb.

Cortex 15 1835–47. https://doi.org/10.1016/j.jns.2011.

10.020.

Description of PD Phonation in Terms of EEG-Related Frequency Bands

233