Information Technology Reference
In-Depth Information
3.3
ERPs, Delay and Musical Interaction
Now that the averaging method for ERP de-noising has been explained, it should be
clear that these sorts of techniques inherently involve a delay. Averaging-based
ERP detection accuracy improves as the number of trials increases. For example, if
a user is asked to choose from an array of (for the sake of argument) 10 different
visual items, each
flashing individually but at different times, it would be usual to
require a minimum of 10 trials (
(flashes) per item to be able to create a suitable
averaged signal for comparing amplitude. Given an inter-stimulus interval of
100 ms between
flashes, there would be 10
flashes multiplied by 10 individual
presentations multiplied by 100 ms
a total test-time of 10 s before the detection
algorithm could assess which of the 10
final averaged chunks contained the highest
peak.
Depending on the skill/experience of the participant, the inter-stimulus interval
might be reduced to 60 ms, and the number of trials down to 7. This would reduce
the time taken to detect the user
s choice down to just over 4 s. However it would
be less accurate and possibly require greater concentration to use.
Importantly, types of time delays are not incompatible with certain types of
musical tasks, more speci
'
cally those types of tasks that are common to creating
music with technology. It is only in the last 25 years that electronic and computer
music has become a predominantly real-time activity, and many high quality pro-
cessing techniques are still time consuming (time domain convolution for example).
Furthermore, many crucial aspects of musical interaction do not require real-time
control. For example, composition can often be an
activity, requiring time
for consideration and planning. In these cases, P300 ERP detection delay times are
not a signi
'
of
ine
'
cant issue.
In addition, as P300 ERP approaches provide direct, time-tagged information
about cognition of sensory events, it can be used as a passive aid to the composition
process. That is to say, the oddball response might feasibly be used as a measure of
novelty given the right presentation paradigm. For example, it is common for
composers and recording artists to listen to complete performances/potential ver-
sions multiple times during production. If EEG recordings could be easily taken
during these listening sessions, EEG responses to musical events could be used to
determine how unexpected and/or attention grabbing each musical event was at the
level of milliseconds. This approach was used in my Audiovisual Composition
Braindrop , mentioned later in this chapter.
The holy grail of ERP detection is what is referred to as
'
'
. This
means that the ERP signal can be effectively detected and separated from the
background EEG noise immediately, pointing directly to the element in the auditory
or visual stream which caused the attentional shift without the need for averaging
multiple tests/trials. These approaches are becoming more possible through
machine learning, and we report results of our research in this area at the end of the
chapter.
Single Trial
 
Search WWH ::




Custom Search