Content uploaded by Roy Francis Navea
Author content
All content in this area was uploaded by Roy Francis Navea on Oct 14, 2015
Content may be subject to copyright.
Beta/Alpha Power Ratio and Alpha Asymmetry
Characterization of EEG Signals due to Musical
Tone Stimulation
Roy Francis Navea
Electronics and Communications Engineering Department
De La Salle University - Manila
Email: roy.navea@dlsu.edu.ph
Elmer Dadios
Manufacturing Engineering and Management Department
De La Salle University - Manila
Email: elmer.dadios@dlsu.edu.ph
Abstract—The complexity of EEG signals make it difficult to
analyze and understand hence a common method is to create
a mathematical model from which approximate measurable
quantities can be derived. Statistical approaches are commonly
used to describe stochastic systems that produce random data
sets. This study aims to characterize EEG signals due to musical
tone stimulation. The characterization is based on the statistical
measures of the power spectrum in the beta and alpha frequency
range. Measures of central tendencies were considered as well as
the alpha asymmetric properties of the left and right hemispheres
of the brain during stimulation. The EEG signals were obtained
using a 14-node neuro-headset from Emotiv. For an initial study,
seventeen volunteers joined the experiment producing 17, 102,
51 and 204 data samples for the Baseline, C, F and G, and
s-Baseline segments, respectively. Data pre-processing, filtering
and power spectrum feature extraction were performed to gather
enough information for analysis. Results show that among the
considered statistical measures, the skewness and kurtosis of the
power spectrum were found to be significant in delineating the
different segments of the audio stimuli. The inverse relationship
between the beta and alpha waves was observed. During active
segments of the audio stimulus, alpha power decreases and
beta power increases and vice versa. Using the difference score
method (DSM) and with the assumption that alpha waves are
inversely proportional to activity, high DSM values were observed
during the active segments of the stimuli thus implying higher
left hemisphere activation. On the contrary, low DSM values
were observed during the inactive segments of the stimuli thus
implying a higher right hemisphere activation. A decreasing DSM
trend was observed from the intermediate node between the
frontal polar site and frontal node to the parietal node. This
is indicative of more frontal activation towards a less activation
on the parietal region.
I. INTRODUCTION
Music is commonly used for entertainment and relaxation.
It is observable that it can put a person in different emotional
states. It can even inspire just like in the life of Albert Einstein.
Music helped him to think about his theories [1] which are
very present in modern technologies. Music is the art and
science that incorporates the arrangement of tones to form
a tune, a good melody and eventually, a musical piece. Music
perception and appreciation varies from person to person as its
melody and rhythm maybe pleasing to one but not for the other
[2]. People perceive music through the auditory system. The
sound transmitted to the inner ear is decomposed according to
the frequency spectrum of sounds. The orderly arrangement of
low to higher frequencies is mapped onto the brain much like
the way low to high notes are mapped on a piano keyboard
[3] .
There are three primary sensations associated with a single
sustained, constantly sounding musical tone: pitch, loudness
and timbre. Pitch is the sensation of altitude or height, loudness
is the sensation of strength or intensity and timbre or tone
quality gives the sensation of distinguishing between the
sounds from different kinds of instruments even if their pitch
and loudness were the same [4].
This paper aims to characterize musical tone as they are
perceived by the brain using electroencephalogram (EEG)
signals. The characterization is based on the power spectrum
of the EEG signals. EEG is a record of the oscillations of
the brain electric potential recorded from the electrodes on
the human scalp using an EEG capturing device [5]. EEG can
reflect the electric potentials of the cerebral activity along with
the activities of the brain. It is objective and sensitive. EEG is
often labeled according to the spectrum of frequencies: delta
waves (0.5-4Hz), theta waves (4-8Hz), alpha waves (8-13 Hz),
beta waves (13-30 Hz), and gamma waves (roughly greater
than 30 Hz) [8]–[10]. This study looked at the possibility of
using the beta/alpha power ratio and alpha asymmetry [6] in
characterizing musical tones in the brain.
II. AUD IO ST IM UL US A ND EEG DATA SET
A. Audio Stimulus
The audio stimulus piece is composed of rests (silence)and
notes (pitch). For an initial study, the notes used followed the
I –IV –V –I major key chord pattern [7]. The notes were played
in the key of C. From the major scale of C (C, D, E, F, G, A,
B), the notes that follow the chord pattern are C, F, G and C.
The piece of the audio stimulus is shown in Fig. 1. The piece
was converted into a 340-second audio file which served as
the stimulus.
Fig. 1. Audio Stimulus Piece
B. Experimental Procedures
Multiple channel EEG data were obtained from the 14-
channels available in the EMOTIV EPOC neuro-headset which
is compliant to the International 10-20 system. The available
channels are AF3, F7, F3, FC5, T7, P7, O1, O2, P8, T8, FC6,
F4, F8 and AF4. The letters F, T, P, C and O stands for frontal,
temporal, parietal, central and occipital lobes, respectively. AF
stands for the lobe intermediate between the frontal polar sites
(Fp) and F while FC is between F and C. The raw data is stored
in a .edf file, converted into a .csv file and then processed in
Matlab.
Seventeen undergraduate students, ages 18 to 21, were
invited to participate in the experiment. The experiment was
conducted in an acoustically prepared room. The participants
used earphones to minimize environmental noise. They lis-
tened to the audio stimulus and their EEG response was
recorded using the Emotiv EPOC neuro-headset. They were
in a relaxed sitting position with eyes closed [13] in a dimly
lighted room to minimize ocular artifacts. The electrodes were
checked to lit from yellow to green in the Emotiv SDK user
interface. The synchronization between the EEG recording
and the audio stimulus was observed using the Problem Steps
Recorder. This made sure that the occurrence of the tone on
the EEG signal is validated in terms of the time when it was
played from the audio stimulus. The timing table in Table I
was used as the basis for locating the notes.
TABLE I
AUD IO STIMULUS TIMING TABL E
The baseline was established during the first 3 minutes of
the audio stimulus [14]. A secondary baseline (s-baseline) was
considered just after when a note was played. The samples are
also mapped to easily locate the notes in terms of the number
of samples. A total of 17 data samples are available for the
Baseline, 102 for C, 51 for both F and G, and 204 for the
s-baseline. The numbers are unequal because of the sequence
of the key chord pattern used.
III. PARAMETERS
A. Beta / Alpha Ratio
This paper focuses on two EEG frequency spectra, the alpha
and beta waves. The ratio is based on the one-sided power
spectrum of the two waves defined by
¯
Pk=(1
N2|X(0)|2,for k= 0.
2
N2|X(k)|2,for k= 1,2, ..., N/2.(1)
where X(k)is the Fourier transform of the time-domain signal
and Nis the number of samples.
High alpha activity is correlated to brain inactivation while
low alpha activity is for brain activation. On the other hand,
Beta waves are associated with the active state of mind.
Alpha activity is observed over the parietal and occipital
lobes while Beta activity is observed in the frontal cortex and
over other areas during intense focused mental activity [11].
To state the relation between alpha and beta waves during
cortical activation, an increase in alpha activity brings about a
decrease in beta activity, and vice versa [6]. This implies that
the beta/alpha ratio increases during activation and decreases
during inactivation. The beta/alpha ratio could therefore be an
interesting parameter to characterize the brain’s response to
musical tones.
B. Alpha Asymmetry
Many parts of the brain process music and it can be
interpreted logically or intuitively. The left hemisphere is
apparently very important for musical abilities which share
properties with speech, such as temporal order, duration,
simultaneity and rhythm. However, the right hemisphere is
very important in many other aspects including the perception
of loudness, timbre, intonation and the expression of emotion
[12].
To define the relative activities of the right and left hemi-
spheres of the brain, the difference score method (DSM )
is used [14]. The DSM is obtained by taking the difference
between the natural logarithms of alpha power response of the
right and left hemispheres of the brain. This can be described
as
ln(R)−ln(L) = ln(R
L)(2)
The difference score results to a simple unidimensional
scale representing the relative activity of the right and left
hemispheres of the brain. With an assumption that alpha waves
are inversely related to activity, higher scores of DSM indicate
relatively greater left frontal activity [14].
IV. DATA PROCESSING
A. Passband Filtering and Normalization
Passband filtering is the selection method by which a
specific range of frequencies can be extracted. Since this paper
deals with the alpha (8-13 Hz) and beta (12-30 Hz) waves
only, the influence of EOG signals which are dominant below
4Hz, ECG at around 1.2Hz, EMG above 30Hz and power
lines at 60Hz has already been reduced much by implementing
a passband filter. The two groups of data obtained from
the filtering method were segmented according to baseline
(EEG in the relaxed state), s-baseline (2-second EEG baseline
before/after a note) and C, F, G (EEG when the respective
notes are played). Each segment per channel was normalized
using the zscore function in eeglab [15].
B. Spectral Estimation and Windowing
The segments per channel were windowed using the Ham-
ming window defined by
whm = 0.54 −0.46cos(2πn
N−1),0≤n≤N−1(3)
This is to reduce spectral leakage due to amplitude disconti-
nuities and aperiodicity in the time-domain [16]. The power
spectrum was obtained using Eq. (1). The baseline signal
contains 23,040 samples while the other segments have 256
samples.
C. Beta / Alpha Power Spectrum Statistical Characteristics
The Beta/Alpha power ratio was characterized in terms of
mean, variance, standard deviation, skewness, kurtosis and
median. These are important measures of central tendency.
The first moment is the mean which is defined by
x=
N
P
1
xi
N(4)
The second moment is the variance defined by
σ2=
N
P
j−1
(xj−x)2
N−1(5)
which is a measure of the spread about the central value.
Square root of the variance is the standard deviation. Higher
moments include skewness and kurtosis. The skewness is a
measure of the asymmetry of the distribution [17]. It is defined
as
skewness =1
N
N
X
j−1
[(xj−x)
σ]3(6)
If skewness is greater than zero (right skewed or positive
skewness), most of the values are concentrated on the left of
the mean with extreme values to the right. If it is less than
zero (left skewed or negative skewness), most of the values
are concentrated on the right of the mean with extreme values
to the left. If skewness is equal to zero, the distribution is
symmetrical around the mean. To be significant, the skewness
should be outside the range of its standard error (SES) which
is defined by
SES =s6n(n−1)
(n−2)(n+ 1)(n+ 3) (7)
where nis the size of the sample. If it is inside the range, a
normal distribution is assumed.
Kurtosis is the measure of the flatness or peakedness of the
distribution [17]. Formally, it is defined as
kurtosis =1
N
N
X
j−1
[(xj−x)
σ]4(8)
Kurtosis greater than 3 (Leptokurtic distribution) indicates that
the values are concentrated around the mean with thicker
tails. This is sharper than a normal distribution with high
probability of extreme values. Kurtosis less than 3 (Platykurtic
distribution) shows a flatter curve as compared to a normal
distribution. The probability for extreme values is less than a
normal distribution, and the they are wider spread around the
mean. Kurtosis equal to 3 (Mesokurtic distribution) is a normal
distribution. For significance, kurtosis should be outside the
range of its standard error (SEK) defined by
SEK = 2(SE S)s(n2−1)
(n−3)(n+ 5) (9)
where nis the sample size. If it is inside the range, a normal
distribution is assumed.
The median is the value of x for which larger and smaller
values of a x are equally probable. Formally for even N, the
median is
0.5(xN/2+xN/2+1)(10)
For odd N, the median is
x(N+1)/2(11)
Parametric tests were used to determine if there is a sig-
nificant difference between the beta and alpha power spectra
mean and the variance at each node using t-test and F-test
statistics, respectively.
D. Feature Data Reliability
The power spectrum vectors of the alpha and beta bands
were tested for reliability using Cronbach’s alpha [18], [19].
This is a statistic measure of internal consistency or reliability
described by
α=Nc
(v+ (N−1)c)(12)
where Nis the number of samples, cis the average inter-
sample covariance among the samples and vis the average
variance.
Acceptable Cronbach’s alpha should be greater than 0.8. If
it falls below 0.2, the data is not reliable at all [18]. In Table
TABLE II
INT ERN AL CONSISTENCY OF BETA AN D ALP HA POW ER SPE CT RUM
II, the internal consistency of the power spectrum vectors of
the alpha and beta bands are shown.
Higher alpha values were observed in the Beta band as
compared to the Alpha band. Nonetheless, an acceptable
reliability is observed among the two bands as the Cronbach’s
alpha values are found to be greater than 0.8.
V. RE SU LTS
The statistical characteristics of the beta and alpha power
ratios were graphed according to the 14 nodes of the EEG
neuro-headset as shown in Fig. 2. Among the six parameters,
skewness and kurtosis were found to significantly delineate
baseline, s-baseline, C, F and G. Other measures were found to
be of no significant difference from each other. Comparing the
alpha and beta power spectra for each node in terms of mean
and variance resulted to a decision that they are significantly
different.
(a) AF3 Node
(b) F7 Node
(c) F8 Node
(d) FC5 Node
(e) FC6 Node
(f) O1 Node
(g) O2 Node
(h) P7 Node
(i) P8 Node
(j) T7 Node
(k) T8 Node
Fig. 2. Beta / Alpha Power Ratio Characteristics
Skewness can be right or left. Based on the results shown
in Table III, the baseline power ratio for all nodes is left
skewed. Most of the s-baseline, C, F and G are right skewed.
Those with significant skewness, with values beyond ±0.1522
for 256 samples and with values beyond ±0.0161 for 23,040
samples, are underlined.
TABLE III
SKEWNESS AND KURTO SI S SIGN IFI CAN CE P ER NOD E
All of the power ratio of each stimulus from the nodes have
significant kurtosis and most of the responses are platykurtic.
Significant kurtosis is beyond ±0.3223 for 256 samples and
±0.0323 for 23,040 samples.
The beta and alpha power ratio was observed during the
inactive and active segments of the recorded EEG. The relation
was expressed in dB and graphed as shown in Fig. 3.
Fig. 3. Beta / Alpha Ratio in dB
Results show that during the inactive segments (Baseline
and s-baseline), the ratio was found to be lower as compared
to the segments where the C, F and G tones were played.
Low dB value indicates that there is a higher alpha activity as
compared to beta activity. It was also observed that when the
tones are played, the dB value increases which implies that
there is a decrease in the alpha activity and increase in the
beta activity.
Baseline and s-baseline segments are of different dB values
wherein the baseline dB value is less than the s-baseline dB
value. This is a possible indication of the unrest in the EEG
activity after being stimulated. The dB value of the s-baseline
is not that far from the dB value of when the tones are played.
Hence, it is still an indication of an increase in alpha activity
and decrease in beta activity as the stimulus fades out and
the EEG signal returns to its stable state. Since there is only
a 2-second interval between the s-baseline and the tones, the
dB value of the s-baseline did not went down equal to the
baseline because of the possibility of there is insufficient time
to make the EEG behavior relax again.
The alpha wave asymmetry between the right and left
hemispheres of the brain was observed. Even-numbered nodes
are located on the right hemisphere and odd-numbered nodes
are on the left hemisphere. Using the DSM, it was found out
that during baseline sessions, the DSM value of the baseline is
less than the DSM of s-baseline, C, F and G segments. Results
are shown in Table IV.
TABLE IV
ALP HA WAVE ASYM MET RY
With the assumption that alpha waves are inversely propor-
tional to activity, high DSM values imply higher left hemi-
sphere activation. For s-baseline, C, F and G, high activation
on the left hemisphere was observed while low activation on
the right hemisphere. Opposite observations were seen for the
baseline.
In Fig. 4, a decreasing DSM trend was observed from the
intermediate node between the frontal polar site (Fp) and
frontal node (F) to the parietal node (AF to P nodes).
Fig. 4. DSM Trend from AF to P node
This is indicative of more frontal activation towards a less
activation on the parietal region. However, from Table IV, an
increased activation in the occipital region was observed.
VI. CONCLUSION
The beta and alpha power spectrum of the EEG signals
during musical tone stimulation was statistically characterized.
Measures of central tendencies were considered as well as the
alpha asymmetric properties of the left and right hemispheres
of the brain. Among the considered statistical measures, the
skewness and kurtosis of the power spectrum were found to be
significant in delineating the different segments of the audio
stimuli. It was found that there is a significant difference
between the beta and alpha power spectrum in each EEG node.
The inverse relationship between the beta and alpha waves
was observed. During active segments, alpha decreases and
beta power increases while during the inactive segments alpha
increases and beta power decreases.
Alpha asymmetry test using DSM was performed. High
DSM values were observed during the active segments of
the stimuli thus implying higher left hemisphere activation.
On the contrary, low DSM values were seen during the
inactive segments of the stimuli thus implying a higher right
hemisphere activation.
REFERENCES
[1] B. Foster. Einstein and his love of Music. Physics World, IOP Publishing,
Bristol, UK, 2005
[2] G. Vecchiato, J. Toppi, L. Astolfi, D. Mattia, P. Malerba, A. Scorpecci,
P. Marsella,. F. Babiloni. Investigation on the pleasantness of music
perception in monolateral and bilateral cochlear implant users by us-
ing neuroelectrical source imaging: A pilot study. Annual International
Conference of Eng’g in Medicine and Biology Society, pp. 8110-8113,
2011
[3] W.J. Cromie. How your brain listens to Music. the Harvard University
Gazette, President and Fellows of Harvard College, 1998
[4] J.G. Roederer. The Physics and Psychophysics of Music: An Introduction,
4th Ed. Springer, p. 4, 2008.
[5] P.L. Nunez, R. Srinivasan. Electric Fields of the Brain: The Neurophysics
of EEG. Oxford University Press, New York, 2006
[6] Q. Dong, Y. Li, B. Hu, Q. Liu, X. Li, L. Liu. A Solution on Ubiquitous
EEG-Based Biofeedback Music Theraphy. 5th International Conference
on Pervasive Computing and Applications, 2010
[7] E. Estrella. Playing Chord Patterns. [Online]. Available:
http://musiced.about.com/od/songwriting101/a/chordpatterns3.htm.
[8] X. Li, B. Hu, T. Zhu, J. Yan, F. Zheng. Towards Affective Learning with
an EEG Approach. 1st ACM International Workshop on Multimedia Tech-
nologies for Distance Leaning, ACM Multimedia 2009,Beijing, P.R.China
[9] J.R. Hughes. Gamma, fast and ultrafast waves of the brain, their
relationships with epilepsy and behavior. Epilepsy Behavior, 2008
[10] S. Sanei, J.A. Chambers. EEG Signal Processing. John Wiley and Sons
Ltd, England, 2007
[11] E.R. Kandel, J.H. Schwartz, T.M. Jessel. Principles of Neural Science,
4th Ed.. Mc Graw Hill, New York, 2000
[12] A. Gates, J.L. Bradshaw. The role of cerebral hemispheres in music.
Brain and Language, vol. 4, iss. 3, pp. 403-431, 1977
[13] F.R. On, R. Jailini, H. Norhazman, N.M. Zaini. Binaural Beat Effect
on Brainwaves based on EEG. 9th International Colloquium on Signal
Processing and its Applications, pp. 339-343, 2013
[14] J. Allen, J. Coan, M. Nazarian. Issues and assumptions on the road from
raw signals to metrics of frontal EEG asymmetry in emotion . Biological
Psychology, v.67, pp. 183-218, 2004
[15] EEGLAB (n.d.)[Online]. Available: http://sccn.ucsd.edu/eeglab/
[16] L. Tan, J. Jiang. Digital Signal Processing Fundamentals and Applica-
tions, 2nd ed. MA, USA: Elsevier Inc., 2013.
[17] H. Berendsen. A Student’s Guide to Data and Error Analysis Cambridge
University Press, 2011
[18] A.J. Tomarken, R.J. Davidson, R.E. Wheeler, L. Kinney. Psychometric
properties of resting anterior EEG asymmetry: temporal stability and
internal consistency Psychophysiology. 1992 Sep; 29(5):576-92
[19] M. Vuga, N.A. Fox, J.F. Cohn, C.J. George, R.M. Levenstein, M. Ko-
vacs. Long-term stability of frontal electroencephalographic asymmetry
in adults with a history of depression and controls International Journal
of Psychophysiology 59 (2006) 107-115
