[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: perceptual learning

Spectral shape perception is in my personal opinion one of
the most promising inroads towards the analysis of perceptual
learning above the "microscopic" level. Indeed I recommend
the work of David Green and Ward Drennan (and a few others)
on spectral profile analysis and the learning aspects of that,
and would like to take this opportunity to plead once more
for much more follow-up work in that direction. Not only for a
better understanding and improvement of cochlear implants and
other hearing aids, but also to learn more about perceptual
and learning issues for blind people accessing purely visual
information through sound representations. For instance, see
yesterday's online article at


which, I'm glad to read, warns for the steep learning curve
in (perceptual) learning for this cross-modal approach. The
technology is now largely there, but our understanding of the 
perceptual and learning issues is not at all at a comparable
level. This is unsatisfactory. The consequence is that we are
working in parallel on exploration through blind volunteers
while further developing the technology, but we largely lack
an understanding of what the brain can learn to make of auditory
images, what the role of brain plasticity and a critical age
would be, or how we should devise a training program for best
and quickest results. This would not be a luxury, because a
steep learning curve can be quite frustrating. Some blind people
nowadays use this technology to hear online stock charts and
the like, but that is trivial as compared to what the technology
will allow for if "only" human perception and learning can be
brought to match what's there from a technical perspective (even
after we consider the basic limitations represented by JNDs,
critical bands and forward and backward masking).

I'd like to see spectral profile learning and analysis applied
to the perception and recognition of for instance various more
or less geometric shapes presented "spectrographically" in sound.
It is easy to parameterize such shapes and sets of shapes for
quantitative studies. This will allow for a scientifically decent
methodology that can systematically extend the work on auditory
profile analysis towards higher levels of human abstraction with
respect to the information content in the sound. Speech and music
seem less suited to that (most people except very young childern
have already mastered at least one language), while many other
classes of sounds (the usual beeps and noises that have been
studied so extensively in the past century) have in my view no
obvious route from the "microscopic" to any worthwhile higher
levels of abstraction in human information processing.

I hope my arguments are not too controversial, because that could
render them counterproductive.

Best wishes,

Peter Meijer

Seeing with Sound - The vOICe