The Effect That an Auditory Distraction with Differing Levels of
Intensity Have on a Visual P300 Speller While Utilizing Low Fidelity
Equipment: Alongside the Development of a Taxonomy
Patrick Schemrbi
a
, Mariusz Pelc
b
and Jixin Ma
c
Department of Computer and Information Systems, University of Greenwich, Greenwich, London, U.K.
Keywords: Brain-Computer Interface (BCI), Electroencephalography (EEG), Event-Related Potential (ERP), P300
Speller, Distractions, Taxonomy.
Abstract: In this paper, we investigate the effect that an auditory distraction with differing levels of intensity has on the
signal of a visual P300 Speller in terms of accuracy, amplitude, latency, user preference, signal morphology,
and overall signal quality. This work is based on the P300 speller BCI (oddball) paradigm and the xDAWN
algorithm, with ten healthy subjects; while using a non-invasive Brain-Computer Interface (BCI) based on
low fidelity electroencephalographic (EEG) equipment. Our results suggest that the accuracy was best for the
no music (M0), followed by music at 90% (M90), music at 60% (M60) and last music at 30% (M30), which
results were in identical order to the subjects' preferences. In addition, the amplitude did not show any
statistical significance in all scenarios while the latency exhibited a minor statistical difference. This work is
part of a larger EEG based project where we are introducing different categories of distractions that are being
considered alongside the development of a taxonomy. These results should give some insight into the
practicability of the current P300 speller to be used for real-world applications.
1 INTRODUCTION
In this paper, we analyze the effect that an auditory
distraction, explicitly that of digital music, with
different levels of intensity in regards to volume (off,
low, mid, high) have on the accuracy, amplitude,
latency, user preference, signal morphology, and
overall signal quality. Our research makes use of non-
invasive Brain-Computer Interface (BCI) on the basis
of Electroencephalography (EEG). The work
presented here is part of a larger EEG based project
and in continuation of our previous papers (Schembri
et al., 2017); (Schembri et al., 2018); (Schembri et al.,
2018); (Schembri et al., 2019).
Event-related potentials (ERPs) are slow voltage
fluctuations or electrical potential shifts recorded
from the nervous system. These are time-locked to
perceptual events following a presentation of a
stimulus being either cognitive, sensor or motor
stimuli. The simplest paradigm for eliciting an ERP is
by focusing attention on the target stimuli (occurs
a
https://orcid.org/0000-0002-7808-5871
b
https://orcid.org/0000-0003-2818-1010
c
https://orcid.org/0000-0001-7458-7412
infrequently) embedded randomly in an array of non-
targets (occurs frequently). The methodology used
derives from the oddball paradigm; first used in ERPs
by Nancy, Kenneth and Steven (Squires et al., 1975),
where the subject is asked to distinguish between a
common stimulus (non-target) and a rare stimulus
(target). The target stimuli elicit one of the most
renowned ERP components known as P300, which is
an exogenous and spontaneous component and was
first described by Sutton (Sutton et al., 1965). The
name is derived from the fact that it is a positive wave
that appears around 300ms after the target stimulus.
Unless otherwise noted herein, the term P300 (P3)
will always refer to a visual P300b (P3b) which is
elicited by task-relevant stimuli in the centro-parietal.
BCI research and development has been
predominantly focused on speed and accuracy of the
BCI application but has been wanting in usability,
such as the environment in which it is being used. In
fact, many BCI applications and experiments were
and are still being performed in laboratory settings
50
Schemrbi, P., Pelc, M. and Ma, J.
The Effect That an Auditory Distraction with Differing Levels of Intensity Have on a Visual P300 Speller While Utilizing Low Fidelity Equipment: Alongside the Development of a Taxonomy.
DOI: 10.5220/0008065200500058
In Proceedings of the 3rd International Conference on Computer-Human Interaction Research and Applications (CHIRA 2019), pages 50-58
ISBN: 978-989-758-376-6
Copyright
c
2019 by SCITEPRESS Science and Technology Publications, Lda. All rights reserved
with unrealistic conditions, where the subject sits in a
sound-attenuated room without any distractions
(Kam et al., 2019) (Bradford et al., 2018). Only a few
research papers such as (Nam et al., 2010) and
(Valentin et al., 2019) focus on real-world contexts,
however they were either using medical grade
equipment (Oliveira et al., 2016) and/or focusing on
auditory ERPs (Zink et al., 2016).
The need for this study originated to broaden the
utilization of this technology for both healthy subjects
and especially to those individuals with severe
neuromuscular disabilities (Sellers, et al., 2006), by
providing a solution based on low fidelity equipment
which is assessed outside lab conditions, and into
noisy environments. Our null hypothesis based on
preceding related and tantamount medical grade
research are that this type of distraction does not show
any statistically significant effect on accuracy, task
performance, amplitude, or latency.
In this work, we report a study where ten healthy
subjects used Farwell and Donchin P300 speller
paradigm in conjunction with the xDAWN algorithm
(Rivet et al., 2009) while utilizing low fidelity
equipment. The subjects were asked to communicate
five alphanumeric characters, referred to as symbols,
in each of the four separate scenarios i.e. off, low, mid
and high volume. The main goal for this study was to
methodically investigate the usability of a P300 BCI
system, explicitly that of a P300 Speller, in a specific
context. Empirical experiments were conducted to
assess how environmental factors such as music, with
different levels of intensity, affect the task
performance and quality of P300 component. This
work is part of a larger EEG based project where we
are introducing different categories of distractions
which are being considered alongside the
development of taxonomy as introduced in Figure 1.
This paper is structured as follows: the equipment,
participants and experimental procedures are
described in Section 2. The offline and online ERP
results are presented in Section 3. Conclusions and
future work are given in Section 4.
Figure 1: Development of extendable Hierarchal Taxonomy
of Distractions for P300b.
2 METHODOLOGY
The following segment/s of the methodology are the
author's previous work as referenced above and are
adopted and outlined in the current paper for readers’
convenience.
2.1 Low Fidelity Hardware
The work reported herein is based on an OpenBCI 32-
bit board (called Cyton) connected with an Electro-
Cap using the international 10/20 system for scalp
electrode placement in the context of EEG
experiments. The Cyton board’s microcontroller is
the PIC32MX250F128B with a 32-bit processor and
a maximum speed of 50MHz; storage of 32KB of
memory and is Arduino compatible. The board uses
the ADS1299 IC developed by Texas Instruments,
which is an 8-Channel, 24-Bit, simultaneous
sampling delta-sigma, Analogue-to-Digital Converter
used for biopotential measurements. The system
comes with a pre-programmed USB dongle for
wireless communication which communicates with
the low-cost RFDuino RFD22301 microcontroller
built on the Cyton board. An additional feature which
is included in the board is a 3-axis accelerometer from
ST with model LIS3DH. This can prove to be quite
useful; such as, for sensing a change in orientation of
the head or sensing rough motion. A more thorough
explanation of the hardware components of the Cyton
board can be found in our previous paper (Schembri
et al., 2017). The Electro-Cap being used in our
experiments has the fabric which is made from elastic
spandex and has recessed pure tin wet electrodes
directly attached to the fabric. The term wet
electrodes type implies that the use of an electrolyte
gel is required to make effective contact with the
scalp otherwise it may result in impedance instability.
A pair of Creative Labs SBS 15 speakers were
used to output the three levels of music. The speakers
have a nominal (RMS) output power of 5 Watt per
speaker, a frequency response of 90Hz – 20,000Hz
and a signal to noise ratio of 90dB.
2.2 Participants
We enlisted a total of N = 10 healthy subjects, seven
males and three females, aged 29-38 (M = 33.8)
which voluntarily participated in this study. Nine out
of the ten subjects’ native language was Maltese and
the tenth subject’s native language was English. All
subjects spoke fluent English and were familiar with
the symbols displayed on the P300 Speller. All
The Effect That an Auditory Distraction with Differing Levels of Intensity Have on a Visual P300 Speller While Utilizing Low Fidelity
Equipment: Alongside the Development of a Taxonomy
51
subjects had previous experience using BCI and
formerly performed P300 speller experiments.
One other subject assisted in the initial testing and
configuration of the equipment; however, he/she did
not take part in the official experiment and hence is
not included in the results.
2.3 Data Acquisition
The EEG signals were sampled at 250Hz, while the
sampling precision was 24-bit. The recordings were
stored anonymously as raw data in OpenVIBE .ov
format. These were later converted to a comma-
separated value (CSV) files for offline analysis. Eight
EEG electrodes were used in different regions of the
scalp according to the International 10-20 System.
The electrode positions C3, Cz, C4, P3, Pz, P4, O1
and O2 were selected as shown in Figure 2. This is
because the spatial amplitude dispersal of the P300
component is symmetric around Cz and its electrical
potential is maximal in the midline region (Cz, Pz)
(Ogura et al., 1995). A referential montage was
selected with the reference electrode being placed on
the left earlobe A1 given that, in general, a mastoid or
earlobe reference will produce a robust P300
response. The right ear lobe A2 was used as ground.
The electrodes are referenced to electrode A1 as
follows: Ch1: C3; Ch2: Cz; Ch3: C4; Ch4: P3; Ch5:
Pz; Ch6: P4; Ch7: O1; Ch8: O2.
Figure 2: Electrode placement following the 10-20 system.
2.4 P300 Speller and xDAWN
In this paper, we make use of Farwell & Donchin
P300 speller, which is based on visual stimuli, in
conjunction with the xDAWN algorithm. Figure 3
depicts what is presented to the subject i.e. a six by
six grid which is made up of thirty-six alphanumeric
characters referred to as symbols. In this
methodology, each row and column of the spelling
grid is augmented in random order and the subject is
asked to distinguish between a common stimulus
(nontarget) and a rare stimulus (target). As a result of
the (target) stimuli, an exogenous and spontaneous
ERP potential known as P300; which is a positive
deviation around 300ms after the stimuli; is evoked in
the brain. The desired symbol is determined and
predicted by the intersection of the (target) row and
column. This prediction entails distinguishing
between non-target i.e. rows/columns stimuli that do
not generate a P300 component and target i.e.
row/column stimuli that generate a P300 component.
In any recorded EEG signal, the P300 component
which has a typical peak potential between 5-10µV,
is embedded and masked by other brain activities
(typical EEG signal +-100µV) such as muscular
and/or ocular artefacts (Schembri et al., 2017) leading
to a very low Signal-to-Noise Ratio (SNR) of the
P300 component. This indicates that it would be very
difficult to detect the target stimuli from a single trial,
which is denoted by a series of augmentation, in
random order, of each of the six rows and six columns
in our matrix (i.e. twelve augmentations per trial). A
popular way to address the limited SNR of EEG is for
each symbol to be spelled numerous consecutive
times and the respective column/row epochs are
averaged over a number of trials, thus canceling
components unrelated to stimulus onset.
The xDAWN process of spatial filtering is (1) a
dimensional reduction method that creates a subset of
pseudo-channels (referred to as output channels) by a
linear combination of the original channels and (2) it
promotes the appealing part of the signal, such as
ERPs, with respect to the noise. This is applied to the
data before performing any classification such as
LDA (Linear Discriminant Analysis) which was used
in this paper. A more thorough explanation of the
xDAWN algorithm can be found in our previous
paper (Schembri et al., 2018) or (Rivet et al., 2009).
Figure 3: BCI “P300 Speller”. The screen as shown to the
subjects with the 3rd row highlighted.
2.5 Experimental Design
In this study, there was one independent variable
manipulated: (a) digital music (off, low, medium and
high) within-subjects variables. In addition, there
CHIRA 2019 - 3rd International Conference on Computer-Human Interaction Research and Applications
52
were several dependent measures used which can be
categorized into three types of dependent variables:
online performance (accuracy), offline performance
(amplitude and latency) and user preference.
2.5.1 Independent Variables
Four-levels of digital music were employed to
represent different real-world scenarios: ‘off’ versus
‘low’ versus ‘medium’ versus ‘high’. The ‘off’ level
cited as M0 represents a lab condition scenario, where
subjects are seated in a sound-attenuated room. The
‘low’ level volume cited as M30 was set at thirty
percent i.e. between 20 and 30dB which simulates
background music. The ‘medium’ level volume cited
as M60 was set at sixty percent i.e. between 50 and
60dB which simulated active listening to a movie.
The ‘high’ level volume cited as M90 was set at
ninety percent i.e. between 80 and 90dB which
simulated disco level music only i.e. no crowd chatter
or noise. This first experiment i.e. M0 was done as a
basis for comparison for M30, M60, and M90.
2.5.2 Dependent Variables
Online Performance (Accuracy): is the number of
correctly spelled symbols over the number of planned
target symbols to be spelled; in our case 5 symbols
which make up the word BRAIN.
Offline Statistics (Amplitude and Latency): P300
Amplitude (μV) is related to the distribution of the
subject’s processing resources assigned to the task. It
is defined as the voltage difference between the
largest positive peak from the baseline within the
P300 latency interval. P300 Latency is considered a
measure of cognitive processing time, generally
between 300-800ms (Stern et al., 2001) poststimulus
i.e. after target stimulus. In simplest terms, it is the
time interval between the onset of the target stimulus
and the peak of the wave.
User Preference: Throughout a questionnaire, the
subjects were asked to rank from one to four, one
being the best and four being the worst, their favorite
usage condition.
2.6 Experimental Procedure
Each subject was invited and attended an induction
session which was aimed to re-educate all subjects on
the P300 speller paradigm and the hardware utilized.
The subjects’ were informed on the following: (1)
they would be performing the experiment in five
unique conditions, in sequence; (a) in the training
phase, in a sound-attenuated room i.e. lab conditions;
(b) M0, (c) M30, (d) M60, (e) M90, as explained in
the independent variable section; (2) the symbols to
spell were “BRAIN” for (1b) to (1e) and fifteen
random symbols for (1a). Any subjects’ query was
answered at this stage. Before the start of the
experiments, each subject was asked to relax for a few
minutes in a seated position. The subject was seated
approximately one meter away from the display. The
researcher and his equipment were situated on the left
side of the subject. The speakers were situated one
meter away and facing the subject at a 15-degree
angle. The experiment was started when the subject
was able to properly perform the task at hand and had
no additional questions. Prior to the start of every
experiment, the impedance of the electrodes was
confirmed to be less than 5K.
The display presented to the subjects is shown in
Figure 3 where 36 symbols presented in a 6x6 matrix.
The target symbol was preceded by a cue i.e. one of
the symbols was highlighted in blue at the beginning
of the symbol run. Each row and column in the matrix
was augmented randomly for 100ms and the delay
between two successive augmentations was 80ms.
This led to an interstimulus interval (ISI) of 180ms.
For each symbol, six rows and six columns were
augmented for fifteen repetitions and there was a
100ms inter-repetition delay and a 300ms inter-trial
period between the end of the trials of one symbol and
the beginning of trials of the next symbol, which
allowed the subject to focus the attention on the next
symbol. At the end of each symbol run, the predicted
symbol was presented with a green cue, which
indicated whether the system predicted the correct
target symbol. The subjects were given a short break
between experiments.
The training phase (1a) consisted of one session
with 15 random symbols by 15 trials each (i.e. 12
flashes of columns/rows per trial * 15 trials = 180
flashes per symbol). The recording of the training
phase took approximately 10 minutes. The M0, M30,
M60 and M90 task experiments consisted of one
session each with the aforementioned conditions and
configurations while spelling the symbols “BRAIN”
consecutively. Similarly to the training phase, each
symbol had fifteen trials each. The recording of each
task lasted approximately 6 minutes. In total, there
were 15 symbols spelled in the training phase and 5
symbols spelled in each task, per subject. Hence due
to the matrix disposition, there were in total 2700
flashes in the training phase, amongst which 450 were
targets; and 3600 flashes in each task (900 * 4 tasks),
amongst which 600 (150 * 4 tasks) were targets.
These values are per subject. The data was stored
anonymously by referring to the subjects as subject1-
10 respectively.
The Effect That an Auditory Distraction with Differing Levels of Intensity Have on a Visual P300 Speller While Utilizing Low Fidelity
Equipment: Alongside the Development of a Taxonomy
53
2.7 Signal Processing
The online system was controlled by OpenViBE 2.0.0
which is a C++ based software platform designed for
real-time processing of biosignal data. The
acquisition server interfaces with the Cyton board
and generates a standardized signal stream that is sent
to the designer which in turn is used to construct and
execute signal processing chains stored inside
scenarios. The signal was obtained via the acquisition
server which does not communicate directly with the
Cyton board. Instead, it provides a specific and
dedicated set of drivers that does this task. The signal
was obtained at a sampling rate of 250Hz with 8 EEG
channels and 3 accelerometer (auxiliary) channels.
The experimental paradigm was controlled by the
OpenViBE designer where a number of scenarios in
the “P300: Basic P300 Speller demo with xDAWN
Spatial Filter” were executed in succession.
In the offline analysis, the following procedure is
done for each M0, M30, M60, and M90. The captured
raw data was converted from the proprietary
OpenVIBE .ov extension to a more commonly used
.csv format using a particular scenario aimed for this
task. The converted data was later imported into
MATLAB R2014a and any unnecessary rows and
columns such as headers and auxiliary data were
removed. Next, we filtered out the data to include the
target stimulations with code (33285); non-target
stimulations (33286); and visual stimulation stop
(32780), which is the start of each flash of row or
column. Subsequently, we had to perform a signal
inversion due to the hardware and driver
implementation. The data (samples and event info)
were later imported into EEGLAB for offline
processing. The first process was to apply a bandpass
filter of 1-20HZ to eliminate the environmental
electrical interference (50Hz or 60Hz), to remove any
signal harmonics and unnecessary frequencies which
are not beneficial in our experiments and to remove
the DC offset. Next, the imported data was used in
ERPLAB which is an add-on of EEGLAB and is
targeted for ERP analysis. We took every event we
wanted to average together and assigned that to a
specific bin via the binlister. This contained an
abstract description of what kinds of event codes go
into a particular bin. In our experiments we have used
the following criteria: “.{33285}{t<50-150>32780}
for the target and “.{33286}{t<50-150>32780}” for
the non-target. This implies that it is time-locked to
the stimuli event 33285 (target) or 33286 (non-target)
and must have the event 32780 that happens 50 to
150ms after the target/non-target event. If this criteria
is met, it is placed in the appropriate BIN; in our case
BIN1 for target and BIN2 for non-target. Next, we
extracted the bin-based epochs via ERPLAB (not the
EEGLAB version) and set the time period from -0.2s
before the stimulus until 0.8s after the stimulus. We
have also used baseline correction (pre) since we
wanted to subtract the average pre-stimulus voltage
from each epoch of data. Next, we passed all channels
epoch data through a moving window peak-to-peak
threshold artifact detection with the voltage threshold
set at 100μV, moving window width at 200ms and
window step at 100ms to remove unwanted signals
such as blinking and moving artifacts. Subsequently,
we averaged our dataset ERPs to produce part of the
results shown in Table 5. Lastly, we performed an
average across ERPsets (Grand Average) to produce
the results in Figure 3 and part of the results in Table
5. The data for Table 5 were generated by the ERP
measurement tool. A more thorough explanation on
segments of the signal processing can be found in our
previous paper (Schembri et al., 2018).
3 RESULTS
In this section, we present several results in relation
to the dependent variables such as a one-way
ANOVA (factorial analysis) to determine the effect
that off, low, mid and high level of volume intensity
have on the online performance (accuracy), offline
statistics (amplitude and latency) and user preference.
In the following tables the labels M0, M30, M60, and
M90 represent “no music - lab condition”, “music at
30%”, “music at 60%” and “music at 90%” volume
respectively. Moreover, M0, M30, M60, and M90
might be interchangeably referred to as BIN1, BIN3,
BIN5, and BIN7 respectively.
3.1 Online Analysis
Following the online experiments, the results
achieved per subject are shown in Table 1 which
depicts the correct symbols predicted out of five (i.e.
symbols BRAIN) and the percentage in parentheses,
rounded to the nearest one, for the accuracy
dependent variable. It must be noted that in an
incorrect symbol prediction, it might be the case that
the column was predicted correctly, whilst the row
was predicted incorrectly or vice versa. For instance,
subject8 had a success rate of 80% in the M30
scenario, with the symbol R predicted as symbol Q
i.e. the row prediction was correct but not the column.
However to avoid ambiguity we have decided to
assume that both row and column prediction were
incorrect when the symbol is predicted incorrectly.
CHIRA 2019 - 3rd International Conference on Computer-Human Interaction Research and Applications
54
Table 1: Symbols spelled (out of 5) and percentage (in
parentheses) for the accuracy dependent variable.
Subject LAB M30 M60 M90
S1
5 (100%) 5 (100%) 4 (80%) 5 (100%)
S2
5 (100%) 5 (100%) 4 (80%) 4 (80%)
S3
5 (100%) 5 (100%) 5 (100%) 5 (100%)
S4
5 (100%) 5 (100%) 5 (100%) 5 (100%)
S5
5 (100%) 5 (100%) 5 (100%) 5 (100%)
S6
5 (100%) 5 (100%) 5 (100%) 5 (100%)
S7
5 (100%) 5 (100%) 5 (100%) 5 (100%)
S8
5 (100%) 4 (80%) 5 (100%) 5 (100%)
S9
5 (100%) 4 (80%) 5 (100%) 5 (100%)
S10
5 (100%) 5 (100%) 5 (100%) 5 (100%)
Grand
Avera
g
e
100%
96%
96%
98%
We have performed a one-way ANOVA which is
based on our independent variable with four
levels/groups (M0, M30, M60, and M90) as presented
in Table 2, to determine if there is a significant
difference between the four means of each group or if
they are all the same. We have chosen to use a 5%
significance level (0.05) denoted as α (alpha) and
rounded all values to the nearest thousandth. Our null
hypothesis (H
0
) states that the means are all equal i.e.
the mean of M0, M30, M60, and M90 are all the
same. Our alternate hypothesis (H
1
) states that at least
two of these means are different.
Table 2: One-way ANOVA test on Accuracy.
Source of
Variation
SS df MS F
P-
value
F
crit
Between
Groups
0.275 3 0.092 0.805 0.499 2.866
Within
Groups
4.1 36 0.114
Total 4.375 39
In the first column we have the source of
variation, where ANOVA carries out an analysis
between groups variation (i.e. M0, M30, M60, M90),
and also carry an analysis of the within groups
variation i.e. the variation within each of our four
groups (refer to Table 1). In the second column, we
have the sum of squares (SS) of the variation, which
is the spread between each individual value and the
mean. The third column is the degrees of freedom (df)
which is the (number of samples – 1). We have four
samples of between groups which gives three and we
have forty samples in total which give thirty-nine.
That allows us to calculate the within-group df which
is total less between groups i.e. a value of thirty-six.
The fifth column we have the mean Square Values
(MS) which is calculated by dividing the SS by the
corresponding df. The sixth column is the F statistic
which is the key statistic where we divide the MS
between groups by the MS within group. Since our F
statistic got a result of 0.805 which is smaller than our
F-critical value (8
th
column), this implies that we
accept the H
0
i.e. that all means are equal and reject
H
1
. Also, by analyzing that the P-value (7
th
column)
which is 0.499 i.e. it is greater than the alpha value of
0.05, so we can also accept H
0
and reject H
1
.
3.2 Offline Statistics
In this section, we process and analyze the averaged
epoch signal of ten subjects in relation to the
independent variable (LAB, M30, M60, and M90).
Figure 3 shows the grand average P300
component for all ten subjects in each scenario which
include all eight channels and an average channel
(AVG). It is comprised of the grand-averaged raw
signals i.e. (5 symbols with 15 trials per symbol); with
(12 flashes of columns/rows per trial); with (10
subjects) i.e. 9000 flashes amongst which 1500 were
targets. In addition figure 3 shows four overlapping
signals, (i) BIN1 - Target for M0 scenario shown in
black (solid for grayscale), (ii) BIN3 - Target for M30
in red (dash-dot) (iii) BIN5 - Target for M60 in blue
(dashed) and (iv) BIN7 - Target for M90 in green
(dotted). To avoid ambiguity and for clarity of the
illustration, we have omitted BIN2, BIN4, BIN6, and
BIN8 which represent the non-target signals.
Table 3 shows the means and standard deviations
in parentheses, for the dependent variables (amplitude
and latency) according to levels of the independent
variable rounded to the nearest hundredth. This data
includes the average of all eight recorded electrodes
throughout the five symbols and is shown per subject
for each BIN1, BIN3, BIN5, and BIN7.
We have performed a one-way ANOVA which is
based on our independent variable with four
levels/groups (Lab, M30, M60, and M90) for our
dependent variables (amplitude and latency) as
presented in Table 4 and Table 5 respectively. In
Table 4 which represents the amplitude, we can see
that the F statistic is 0.723 which is smaller than our
F-critical value of 2.866. In addition, our P-value is
0.545 which is greater than the alpha vale. This
implies that we can accept the null hypothesis (H
0
)
and reject the alternate hypothesis (H
1
). In Table 5
which represents the latency, we can see that the F
statistic is 2.982 which is slightly larger than our F-
critical value of 2.867. In addition, the P-value is
0.044 which is slightly smaller than the alpha value.
This implies that we reject H
0
and accept H
1
. A more
thorough explanation of the one-way ANOVA can be
found in the previous section.
The Effect That an Auditory Distraction with Differing Levels of Intensity Have on a Visual P300 Speller While Utilizing Low Fidelity
Equipment: Alongside the Development of a Taxonomy
55
Table 3: Means and Standard Deviations (in Parentheses) for Two Dependent Measures (Amplitude and Latency).
Subject
LAB
(BIN 1)
M30
(BIN 3)
M60
(BIN 5)
M90
(BIN 7)
Amplitude
(μV)
Latency
(ms)
Amplitude
(μV)
Latency
(ms)
Amplitude
(μV)
Latency
(ms)
Amplitude
(μV)
Latency
(ms)
S1
4.90
(
0.50
)
466.5
(
2.97
)
2.09
(
0.28
)
473.5
(
12.46
)
3.90
(
0.40
)
470.5
(
5.21
)
2.88
(
0.40
)
462.5
(
19.00
)
S2
2.39
(1.14)
437.0
(82.16)
3.24
(0.35)
434.0
(76.58)
4.26
(0.41)
419.5
(54.15)
3.66
(0.83)
416
(66.69)
S3
3.29
(1.68)
431.0
(83.52)
2.18
(2.24)
417.0
(71.61)
4.03
(0.99)
430.0
(85.68)
3.64
(1.42)
423
(78.49)
S4
3.70
(
0.98
)
466.0
(
107.48
)
4.95
(
0.54
)
427.0
(
82.11
)
4.14
(
0.45
)
432.5
(
86.76
)
3.25
(
1.08
)
419.5
(
75.49
)
S5
3.96
(
0.78
)
444.0
(
91.49
)
4.11
(
0.95
)
431.0
(
84.58
)
4.61
(
0.76
)
428.5
(
80.21
)
4.77
(
1.17
)
436.5
(
85.56
)
S6
5.39
(1.64)
483.0
(1.85)
3.63
(2.44)
479.0
(17.73)
4.99
(2.40)
484.5
(3.96)
3.70
(2.12)
495
(2.83)
S7
1.70
(
1.66
)
430.5
(
81.75
)
4.94
(
1.26
)
431.5
(
84.97
)
4.18
(
1.24
)
427.5
(
79.81
)
4.58
(
1.34
)
449.5
(
93.64
)
S8
2.64
(
1.38
)
415.0
(
71.10
)
2.51
(
1.57
)
440.0
(
72.88
)
3.52
(
1.22
)
415.5
(
66.91
)
2.18
(
1.69
)
446.0
(
90.16
)
S9
4.46
(1.94)
439.0
(85.89)
2.50
(1.40)
415.5
(71.66)
2.17
(2.05)
424.0
(76.67)
3.67
(1.89)
411.0
(75.46)
S10
3.58
(0.26)
497.5
(87.40)
4.08
(0.71)
420.5
(80.61)
3.46
(0.61)
420.5
(76.98)
3.51
(0.46)
424.5
(84.43)
Grand
Avera
g
e
3.60
(
1.83
)
440.0
(
85.31
)
3.43
(
1.66
)
429.0
(
80.87
)
3.93
(
1.38
)
427.5
(
79.98
)
3.59
(
1.46
)
431.0
(
81.04
)
Figure 3: Grand average P300 for all 10 subjects in each scenario with all eight channels and an average channel (AVG).
C3
-200 200 400 600
-5
-3.8
-2.5
-1.3
1.3
2.5
3.8
5
Cz
-200 200 400 600
-5
-3.8
-2.5
-1.3
1.3
2.5
3.8
5
C4
-200 200 400 600
-5
-3.8
-2.5
-1.3
1.3
2.5
3.8
5
P3
-200 200 400 600
-5
-3.8
-2.5
-1.3
1.3
2.5
3.8
5
Pz
-200 200 400 600
-5
-3.8
-2.5
-1.3
1.3
2.5
3.8
5
P4
-200 200 400 600
-5
-3.8
-2.5
-1.3
1.3
2.5
3.8
5
O1
-200 200 400 600
-5
-3.8
-2.5
-1.3
1.3
2.5
3.8
5
O2
-200 200 400 600
-5
-3.8
-2.5
-1.3
1.3
2.5
3.8
5
AVG
-200 200 400 600
-5
-3.8
-2.5
-1.3
1.3
2.5
3.8
5
BIN1: Target - Lab
BIN3: Target_Music30
BIN5: Target_Music60
BIN7: Target_Music90
CHIRA 2019 - 3rd International Conference on Computer-Human Interaction Research and Applications
56
Table 4: One-way ANOVA on Amplitude.
Source of
Variation
SS df MS F
P-
value
F
crit
Between
Grou
p
s
2.798 3 0.933 0.723 0.545 2.866
Within
Grou
p
s
46.415 36 1.289
Total
49.212 39
Table 5: One-way ANOVA on Latency.
Source of
Variation
SS df MS F
P-
value
F
crit
Between
Grou
p
s
3236.4 3 1078.8 2.982 0.044 2.867
Within
Groups
13025.6 36 361.822
Total
16262 39
3.3 User Preference
Exactly after the experiments were finished, each
subject was presented with two questionnaires (a) and
(b) to specify their preferred BCI usage condition.
The ranking consisted of maximum weight value of
four (4) as the most desired and minimum weight
value of one (1) for the least desired.
In the first questionnaire, the subjects were
allowed to give the same ranking to different groups
as shown in column (a) of Table 6. Expectedly, the
M0 scenario got the highest ranking, whilst
surprisingly M90 came second, M60 third and M30
last. The frequency analysis grouped by the highest
value of four (4), shows that the M0 was given 100%,
followed by M90 with 50% trailed by M30 and M60
equally at 40%.
In the second questionnaire, the subjects were
asked to give a unique ranking (1-4) to each scenario
as shown in column (b) of Table 6. The results are
similar to those achieved in the questionnaire (a),
where M0 came first, followed by M90, M60 and M30
respectively. The frequency analysis shows that M0
got 40%, followed by M90 at 23%, M60 at 22% and
lastly by M30 at 15%.
Our assumption on why M90 placed second and
M30 placed last in both questionnaires is that, since
the experiments were performed in sequence, the
subjects were staggered by the difference between M0
and M30, while they gradually got accustomed to the
music and hence they were not bothered or distracted
as they initially were. In addition, the results indicated
that the user preference wasn’t affected by the
loudness of the music.
Table 6: User preference for two questionnaires with (a)
allowing the same ranking and (b) unique ranking.
Subject
LAB M30 M60 M90
(
a
)
b
(
a
)
b
(
a
)
b
(
a
)
b
S1
4 4 3 2 3 1 4 3
S2
4 4 4 3 3 1 3 2
S3
4 4 3 1 3 3 3 2
S4
4 4 3 1 3 2 4 3
S5
4 4 4 3 4 2 3 1
S6
4 4 4 1 4 3 4 2
S7
4 4 4 1 4 2 4 3
S8
4 4 2 1 3 2 3 3
S9
4 4 2 1 3 3 4 2
S10
4 4 3 1 4 3 3 2
Total
40 40 32 15 34 22 35 23
4 CONCLUSION
In this study, N = 10 healthy subjects were able to
perform several experiments using Farwell &
Donchin P300 speller in conjunction with the
xDAWN algorithm, with a six by six matrix of
alphanumeric characters, in M0, M30, M60, and M90
environments while utilizing low fidelity equipment.
The goal of this study was to investigate the
usability of a visual P300 Speller, and assess the
extent of effect that an auditory distraction such as
digital music, with varying level of intensities (off,
low, mid, high), have on the user and overall BCI
performance (i.e. the dependent variables: accuracy,
amplitude, latency, and user preference). This study
is part of a larger-based project where we are
introducing different categories of distractions which
are being considered alongside the development of a
taxonomy and give some insight on the practicability
of real-world application of the current P300 speller
with our aforementioned low-cost equipment.
Our null hypothesis based on preceding related
and tantamount medical grade research was that this
type of distraction as elucidated in the independent
variable does not show any statistically significant
effect on the accuracy, amplitude, and latency
dependent variables. The results of a one-way
ANOVA factorial analysis accepts our null
hypothesis for the accuracy and amplitude dependent
variables, however, it was rejected for the latency
dependent variable since there was a minor statistical
significance as shown in the results section.
Non-statistical results show that the dependent
accuracy variable was highest in the M0 (100%) and
surprisingly followed by M90 (98%) trailed by M30
and M60 equally at 96%. Our empirical evidence
The Effect That an Auditory Distraction with Differing Levels of Intensity Have on a Visual P300 Speller While Utilizing Low Fidelity
Equipment: Alongside the Development of a Taxonomy
57
suggests that the subjects got accustomed to the music
in the M90 environment since these were performed
in sequence as explained before. The dependent
variable amplitude was highest in the M60 (M=3.93,
SD=1.38) followed by M0 (M=3.60, SD=1.83), M90
(M=3.59, SD=1.46) and M30 (M=3.43, SD=1.66).
Additionally, the dependent variable latency was
shortest in M60, followed by M30, M90 and finally
M0 as shown in Table 3. It seems that there is no
correlation between amplitude and latency. Lastly,
the user preference evidently shows that all subjects
preferred the M0, followed by M90, M60, and M30 in
both questionnaires as shown in Table 6. This
enforces our previous empirical evidence that the
subjects seem to get acquainted with the music in the
fourth sequential experiment of M90 while they are
staggered by the difference between M0 and M30,
which follow each other. These results also indicated
that the user preference wasn’t affected by the
loudness of the music. Moreover, the signals were
morphological consistent in all four scenarios, even
though they did not yield identical P300 components.
In the future we plan to run the independent
variable levels (M0, M30, M60, M90) experiments in
a randomized order and not sequentially, to avoid the
results being affected by subjects accustomization to
the distraction. Another important point to take into
account in future experiments is the possible impact
of mental fatigue with and without the presence of
distractions during repetitive exercises.
Our main contribution is the comparative
assessment in terms of (a) accuracy, (b) amplitude, (c)
latency and (d) user preference, between the levels of
the independent variable. Our main goal is to provide
insight into the practicability of the current P300
speller to be used in concurrence with several
taxonomized distractions.
In this paper, we have introduced our expandable
hierarchical taxonomy as depicted in Figure 1. This
work is part of a larger EEG based project where we
are introducing different categories of distractions
which are considered alongside the development of
taxonomy while using low fidelity equipment. Our
investigation is concerned with the way in which
different types of distractions (e.g. audio, visual, with
differing intensity/regularity and engagement factor)
translate into a reduction of the signal quality and
amplitude, or any other change/distortion that occurs.
REFERENCES
Bradford, C. J. et al., 2018. Performance of conformable,
dry EEG. 40th Int Conferenc of EMBC, pp. 4957-4960.
Kam, J. W. et al., 2019. Systematic comparison between a
wireless EEG system with dry electrodes and a wired
EEG system with wet electrodes. NeuroImage, 184(1).
Nam, C. S., Li, Y. & Johnson, S., 2010. Evaluation of P300-
Based Brain-Computer Interface in Real-World
Contexts. International Journal of Human-Computer
Interaction, 26(6), pp. 621-637.
Ogura, , Koga , & Shimokochi, , 1995. Recent Advances in
Event-related Brain Potential Research: Proceedings of
the 11th International Conference on Event-related
Potentials (EPIC), Japan, June., 1995. Elsevier.
Oliveira, A. S. et al., 2016. Proposing Metrics for
Benchmarking Novel EEG Technologies Towards
Real-World Measur. Frontiers in Human Neuroscience
Rivet, B., Souloumiac, A., Attina, V. & Gibert, G., 2009.
xDAWN Algorithm to Enhance Evoked Potentials:
Application to BCI. IEEE Transactions On Biomedical
Engineering, 56(8), pp. 2035 - 2043.
Schembri, P., Anthony, R. & Pelc, M., 2017. Detection of
Electroencephalography Artefacts using Low Fidelity
Equipment. Proceedings of the 4th Int. Conference on
Physiological Computing Systems, pp. 65-75.
Schembri, P., Anthony, R. & Pelc, M., 2018. The
Feasibility and Effectiveness of P300 responses using
Low Fidelity Equipment in three Distinctive
Environments. 5th Int. Conf. Physiological Computing.
Schembri, P., Anthony, R. & Pelc, M., 2018. The Viability
and Performance of P300 responses using Low Fidelity
Equipment. 5th Int Conf. on Biomedical Eng. Systems.
Schembri, P., Pelc, M. & Ma, J., 2019. Comparison
between a Passive and Active response task and their
effect on the Amplitude and Latency of the P300
component for Visual Stimuli while using Low Fidelity
Equipment. 41st Annual International Conf of the IEEE
Engineering in Medicine and Biology, EMBC 2019.
Sellers, E. W., Kübler, A. & Donchin, E., 2006. BCI
Research at Univ of Florida Cognitive Psychopysiology
Laboratory: The P300 Speller. IEEE Trans On Neural
Systems Rehabilitation Engineering,14(2),pp. 221-224.
Squires, N., Squires, K. & Hillyard, S., 1975. Two varieties
of long-latency positive waves evoked by unpredictable
auditory stimuli in man. EEG Clinical Neurophysiol,
Stern, , Ray, J. & Quigley, K.S., 2001. Psycho-
physiological Recording. 2nd ed. Oxford University.
Sutton, S., Braren, M., Zubin, J. & John, E. R., 1965.
Evoked-Potential Correlates of Stimulus Uncertainty.
Science, 150(3700), pp. 1187-1188.
Valentin, O. et al., 2019. Validation and Benchmarking of
a Wearable EEG Acquisition Platform for Real-World
Applications. IEEE TRANS ON BIOMEDICAL
CIRCUITS AND SYSTEMS, 13(1), pp. 103-111.
Zink, R., Hunyadi, B., Huffel, S. V. & Vos, M. D., 2016.
Mobile EEG on the bike: disentangling attentional and
physical contributions to auditory attention tasks.
Journal of Neural Engineering, 13(4).
CHIRA 2019 - 3rd International Conference on Computer-Human Interaction Research and Applications
58