Information Technology Reference
In-Depth Information
the same stimuli in order to gauge whether the response is positive or not. This extra
time adds a delay to the signal processing, distancing control away from real-time
musical in
uence.
10.8
EEG Classification and Auditory Stimuli
By the early 2000s, there were several headband-based systems that could play
music from EEG data (Miranda
2001
). The majority of these provided only two
electrodes and very limited tools for interpreting the raw EEG data. Moreover, the
quality of the EEG obtained with these less costly systems did not match the
minimum standards required to implement reliable BCI system. Nevertheless, in
2001, Alexander Duncan, then a PhD candidate working under the guidance of
Eduardo Miranda and Ken Sharman at the University of Glasgow, proposed a
BCMI system based on musical focusing through performing mental tasks whilst
listening to music, alongside EEG pattern classi
cation (Duncan
2001
). Duncan
'
proposed a number of data classi
cation methods for collecting a subject
s EEG
pro
er, which is used for comparative
analysis of EEG readings. This system could effectively be trained to understand the
brain signals of a user so that in practice there was a built-in model to apply
'
best-
le to create an of
ine neural network classi
rules to derive the meaning within the EEG. Here, EEG was extracted through
power spectrum analysis, instead of ERPs. Power spectrum analysis uses Fourier
transformations to observe the amplitudes of EEG frequencies. In this set-up, EEG
generated from external stimuli was analysed by a computer to create classi
t
'
cations
of patterns over multiple trials. Building such a classi
cial
intelligence to create models of expected users responses. A model is built from the
averages of many practice tests of an individual
cation systems used arti
s response to stimuli, which in
effect trains the system. When the system is then engaged in an experiment, it reads
an incoming EEG signal and classi
'
es it against the arti
cial neural network stored
within its memory.
Researchers based at the Interdisciplinary Centre for Computer Music Research
(ICCMR), University of Plymouth implemented this approach in experiments that
combined auditory attention with data classi
cation to analyse features within a
short epoch of post-stimuli EEG. In 2003, Miranda and colleagues reported on three
experiments that investigate methods of producing meaningful EEG, two of which
were deemed suitable for practical musical control. The
first of the two uses the
technique of
active listening
, and the second uses
musical focusing
.
In the
first experiment, small epochs of EEG measured across 128 electrodes
were analysed to determine any difference between the acts of
active listening
(replaying a song in the
minds ear
) and
passive listening
(listening without focus).
Trials were multiplied and looped to build a portfolio of EEG readings. Musical
stimuli consisted of melodic phrases being played over rhythmic patterns. In dif-
ferent trials during a break between melodies, subjects were asked to do three
different things. In the
first trial to replay the tune in their heads, in a second to try
