On Mapping EEG Information into Music - Guide to Brain-Computer Music Interfacing - page 235

To develop sophisticated systems of BCI, control relevant stimuli are required,
and unless using in-the-box methods of analysis and data processing, the appro-
priate means of data acquisition and methods of mapping to a musical engine are
necessary, and this requires expertise.
In 2005, Miranda adopted the approach of designing the musical engine of a
BCMI with suf cial intelligence in order to create sophisticated meaning

from simpler EEG readings. Here, he applied a process known as Hjorth analysis, a
second method of extracting EEG alongside power spectrum analysis. Hjorth
analysis is the extrapolation and measure of time-based features within short
windows of EEG information. These are referred to as the activity, mobility and
complexity within the reading, and measures of each are produced involuntarily as
they lie within overall EEG data. Using these techniques, the BCMI - Piano attempts
to guess the mental state of the user and performs real-time generative piano music
in response, with features based on the techniques of composers such as Beethoven
and Schumann, as discussed in Chap. 3 .
The P300 oddball paradigm, earlier mentioned in relation to auditory stimuli
research, was used by Grierson ( 2008 ) for a BCMI controlled by focusing visual
attention to stimuli displayed on a computer screen (See also Chap. 3 ). The P300
potential was found to contain information relative to visual attention of repetitive
stimuli. In the same manner, as deviations in auditory stimuli were found to trigger
P300 responses (N

cient arti

nen 1990 ) as an automatic response, the P300 could also be
elicited by an unexpected interruption within a repetitive visual pattern. In the case
of P300 spelling devices that allow a user to select letters to form words and
sentences, the deviant information contains the letter the user desires, and as such is
injected with the meaning that a BCI system can knowingly respond to. In the

ää t ä

rst
incarnation of his BCMI, Grierson replaces letters for musical notes for a user to
select via a visual interface.
Over the course of trials, Grierson recorded that four out of

ve subjects were

able to perform subjective decision making, with regard to speci c note selection

and with no training, that were understood by the system 75 % of the time. As ERPs
are dif

cult to detect within EEG, conducting multiple trials improves the reliability
of the system to detect these choices and increases the percentage of success. The
downside is the time lapse introduced from the initial cognitive decision being
made to the end of the trials and the subsequent data processing. Grierson recog-
nises this factor opting for a minimal trial approach in an attempt to link control as
close to cognition as possible. The stimuli in this system presented the names of
note values over three octaves. Each note name was displayed for approximately
50 ms then removed for up to 1,800 ms, in a quasi-random order. A subject was
asked to select a speci c note and count each time it was displayed, generating the

associated ERP information in synchronisation with each display. Experiments
recorded time delays of approximately 12 s, with one subject successfully initiating
control over approximately 7 s with less trials, where total time = ash

time × choices × trials, e.g. 50 ms × 36 × 7 = 12.7 s.