Electroencephalograph (EEG) Signal Feature Extraction For Unspoken-Speech Identification Using EEGLAB

Electroencephalograph (EEG) Signal Feature Extraction For
Unspoken-speech Identification Using EEGLAB
Abstract
In patients with total paralysis, nervous ability to send commands from the
brain to the muscles experiencing distraction. The brain is the only organ that
is still functioning normally. Unspoken-speech becomes a communication
channel important for people with paralysis because it is not requires muscle
activity. The brain is an organ of the body generate electric potential in every
activity. Potential the electricity is measured and recorded as a signal
electroencephalograf (EEG). EEG signal has many characteristics or certain
features that represent the response to brain activity certain. In this Thesis, a
study for extracting the features contained by the EEG signal on activity
unspoken-speech has been carried out. Two participants were instructed
pronounce the vowels / a / / i / / u / / e / / o / without moving articulatory
organs and produce no sound at all. Signal EEG is processed using EEGLAB
and software produce several features such as event related potential (ERP),
ERP image, event related spectral perturbation (ERSP) and inter trial
coherence (ITC). The result of the pronunciation comparison of the letter
/ a / from the two participants indicated that the ERP pattern was similar to
one each other. ERP image shows consistency of deep trial form an ERP
pattern. Meanwhile, ERSP and ITC produce patterns different spectrograms.
Methods
EEG Signal Recording Recording of the EEG signal is carried out in a medical
clinic using Compumedics ProFusion. Installation the electrodes follow the 10-
20 standard with the number of channels as many as 21 pieces. EEG signals
recorded from two people adult male subjects. Subjects were given visual
stimuli on the monitor screen by reading without Vowel presentation sound a
/ i / u / e / o with Blocks model and Sequential according to experiment.
The data that has been recorded are still continuous data. During that time,
the participants carried out activities range from muted pronunciation to
sessions sequential and blocks to non-brain activity raises artifacts. Each
session has an estimated time beginning to end time. Therefore, the
continuous data need to be segmented according to the time span and related
subject activities.
This data segmentation method is quite easy, just stay synchronizes record
time in ProFusion with time on the video camera. Then on continuous data
that is in the form of text searched for event time when the participant starts
pronounce the letters. This is governed by the appearance of the stimulus
whose timing is set with a 2 second delay per letter. After we get the time
data for each event occurs then the sequence of events is arranged in a text
file in order can be read by EEGLAB. EEGLAB delivers easy to read the
arrangement of event data that has been made. Namely by creating a text file
from the application notepad by writing the parameter fields for type, latency
and the duration of each event.
Epoch is EEG data that is limited by time span event based. So after
continuous data is marked each event at a certain time position then the epoch
can extracted from continuous data. The duration of the epoch is set 0.9
seconds, starting from 0.1 seconds before the event until 0.8 seconds later.
Because the data you want to observe, namely about 800 milliseconds after
the stimulation event.
Data Decomposition using ICA
Other research says that EEG signals are recorded on the electrodes does
not only come from the region related brain but a combination of other nerve
cells which are scattered in various other brain areas. This matter causes the
information it wants to extract from the brain response due to certain cognitive
functions being covered up. Independent Component Analysis (ICA) is a
method to separate the desired signal components from mix other EEG signals
using the principle spatial filter.
In this study the ICA algorithm used is Infomax which is integrated with
EEGLAB. Results of research shows that the Infomax algorithm ranks the
highest level of efficiency in separating artifacts and neural signals.
ERP
One of the feature extraction methods that has long been used in EEG
signals is averaging. The basic principle of this method is to amplify signal
patterns that have the same phase at a certain time (time locked) and
attenuate signals with different phases. Averaging is performed on the ICA
decomposition result signal which is the component whose contribution is to
form the largest ERP wave. From approximately 20 trials for each letter, 1
averaging signal pattern is obtained.
PSD
Power spectra density graphs are useful for determining EEG or ERP signal
activity in the frequency domain. The features of the EEG signal are known to
be in the alpha, beta, theta and delta frequency rhythms. Through this
method, it can be seen at which frequency ERP activity is greatest.
Signal Similarity Analysis
This method is done so that the signal that is believed to be a feature is

proven or not. It can be seen the pattern of similarity of two or more signals
by measuring the level of their correlation coefficient. This method is based
on the corr () function in Matlab. The level of correlation is measured on a
scale of 0 to 1. It is worth 0 if the two signals being compared have no
similarity at all, while it is worth 1 if the signals being compared have the
same pattern.
Results
1. In this study, the methods used to process EEG signals are signal
segmentation, visual inspection, averaging and data decomposition
using the ICA algorithm. The four methods have been implemented on
the data using EEGLAB software.
2. The features that have been successfully obtained are event related
potential, ERP image, power spectral density and root mean square.
3. The results of the EEG signal decomposition using ICA show an irregular
pattern and do not produce ERP patterns that respond to stimuli.
4. ERP visual inspection results provide a signal pattern that shows a clear
response to stimuli because the signals from all channels have a similar
pattern.
5. The analyzed EEG signal contains a frequency in alpha rhythm. This can
be seen in the PSD graph where the significant PSD value is in the 10
Hz frequency range.
6. The results of the signal similarity analysis visually show that the ERP of
the five subjects has a similar signal pattern over a time span of 250 ms
to 300 ms.
7. The result of the calculation of the signal similarity coefficient gives a
low value for each comparison. The mean result of all correlation
coefficients is 0.28. B. Suggestions To improve this Thesis research so
that in the future it can get better results, some suggestions that can
be given are: 1. Using EEG hardware with more electrodes, namely more
than 128 channels so that each EEG signal activation can be analyzed
spatially. 2. Increase the number of participants so that the
experimental data becomes more so that the pattern of features
obtained becomes more accurate. 3. Giving multiple kinds of stimuli
such as saying a word or sentence. 4. Time is recorded exactly when
the participant responds to the stimuli. This needs to be done so that
the results of the analysis regarding the response to stimuli become
more accurate. 5. Perform classification or feature recognition processes
so that the Brain Computer Interface system is realized. D

Electroencephalograph (EEG) Signal Feature Extraction For Unspoken-Speech Identification Using EEGLAB

Uploaded by

Copyright:

Available Formats

Electroencephalograph (EEG) Signal Feature Extraction For Unspoken-Speech Identification Using EEGLAB

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Electroencephalograph (EEG) Signal Feature Extraction For Unspoken-Speech Identification Using EEGLAB

Uploaded by

Copyright:

Available Formats

Electroencephalograph (EEG) Signal Feature Extraction For

Unspoken-speech Identification Using EEGLAB

Data Decomposition using ICA

Signal Similarity Analysis

This method is done so that the signal that is believed to be a feature is

You might also like