Biomedical Engineering Reference
In-Depth Information
Sound
FEM
Model
Instrumen
tal
% C
orrect
Classifier
Coding
DCT
Sound
Strategy
Noise
Fig. 11
Block diagram of the automatic instrument recognizer
Ta b l e 6
Results of the automatic instrument identification task for different stimulation modes and
different amounts of neural degeneration
Scores (%)
Node 10 (%)
Node 8 (%)
Node 7 (%)
Monopolar
50
50
50
Bipolar
55.55
66.66
50
Tripolar
72.22
72.22
55.55
classifier on timbral features, which are more relevant for instrument identifaction.
Some instrumental sounds were processed with the described method to obtain the
templates that will be used as reference to perform the classification. To classify
new instrumental sounds, the same process as the one described to compute the
templates is followed. Next, the classifier compares the values generated by the
new instrumental sounds with the stored templates. The instrument template that
minimizes the dissimilarity with the new sound is selected by the class it belongs
to. The similarity measure uses Dynamic Time Warping (DTW) [
18
]. Instrumental
sounds containing a single note (B3 and C4) of a cello, viola, trumpet, trombone, flute
and saxophone were used to compute the templates. The % correct classification was
obtained by processing single notes of the same instruments containing the notes C3
and B4. That means that 12 instrumental sounds were used for the evaluation. Table
6
presents the results of this experiment which serves to illustrate the variability in the
identification of sounds for different stimulation modes and different amounts of the
neural degeneration.
4 Discussion
This manuscript has presented the development of a model of the electrically stimu-
lated cochlea based on FEM. The model has been developed with the goal to assess
individual differences on the peripheral processing of cochlear implant users. For
example the spread of excitation which is believed to contribute to speech intelligi-
bility and sound performance in CI users has been examined in four experiments.
Experiment 1 analyzed the variability in the extracellular voltage and the width
of the spread of excitation for different electrode positions (modiolar, midscalar and
lateral). The model was able to predict that in general modiolar electrode position is
able to produce narrower voltage distribution if there is auditory nerve degeneration.
