cochlear implant/musical instrument identification

[Rebecca Danielle Reich <rreich@xxxxxxx>]

Dear list, [apologies for the long post]

I am a Master's degree student investigating the basic signal processing
of a cochlear implant (specifically, the Clarion CIS strategy) to try to
determine what cues used to identify musical instruments are being lost in
the processing.  The signal processing I am simulating consists of 1 to 8
channels of logarithmically-spaced bandpass filters, full-wave
rectification and smoothing (a moving-average lowpass filter).  I am
creating sinusoids at the frequencies of the center frequencies of the
bandpass filters and summing up the signal to deliver to normal-hearing
listeners over headphones (this simulation is similar to Loizou's
simulation, http://www.utdallas.edu/~loizou/cimplants, but with different
parameters).  I plan on running test audio signals through the simulations
and presenting the output signals to normal-hearing listeners over
headphones.  I would like to investigate more than just the number of
channels required to achieve an acceptable (say, 70%) recognition rate; I
would also like to find out more specifically what information is being
lost to be able to suggest a better processing scheme for music signals.

My question is: what would be the best type of audio signals to use for
this experiment?  I am limiting my signals to solo instruments, and was
thinking along the lines of 5-note ascending/descending scales (to have
note transitions and not isolated tones).  Would synthetic instruments be
cleaner and thus simpler to use than acoustic?  Would a simpler experiment
using only synthesized tones yield more insight?  I am having trouble
correlating the information I would like to obtain with the signals I need
to use in order to obtain that information.  I would appreciate any
suggestions on input signals or experiment design.

Thank you,
Rebecca

Rebecca Reich
Music, Mind and Machine Group
MIT Media Lab
O: 617.253.6469
H: 617.225.9869
C: 617.407.0831