UWE Bristol Engineering showcase 2015

Shrey Hirpara
BEng Electronics Engineering
Project Supervisor
Mokhtar Nibouche
Electronic Ear - Investigation and Implementation of computational
models of a human ear.
Introduction
Automatic speech recognition can be defined as the ability of an independent electronic system to translate a spoken language into a
readable text in real time. It recognition has gained a lot of importance over last decade The need for efficient speech recognition
system has become essential in current day applications. An effective approach to solve this problem would be to model the human
cochlea which will give an insight in to how humans perceive speech. The human cochlea is responsible for converting the sound energy
in to electrical signals which is the inherent reason for hearing. The cochlea model designed in this investigation is based on the Lyon's
Cochlea gram model.
Design Parameters
As seen in Figure 2, 3 inputs and a audio input are added together which pass through the gain then the filterbank
(inside the subsystem). 36 filters are used in the filterbank with one Lowpass filter, 34 Bandpass and one Highpass
filter. They are set up in cascaded parallel configuration as seen in figure 3. The white boxes represent the filters
and the yellow boxes represent the spectrum scopes after each filter to observe its output. All filter outputs are
added together to form the final signal. Two models are designed where one used FIR filters and the other uses
IIR.
Results
Figure 1. Depicts Lyon’s cochlear model.
Figure 2. Shows the designed cochlear
model in Simulink.
As seen in figure 4, the output graph shows all the frequency components of the input signal which are at 1kHz,
10kHz and 15kHz. Simulation using real life sound was also conducted using the ‘Audio Device’ block in figure 2,
where the output showed all the frequency components. Individual filters were also tested with the passband and
stopband frequencies. The filters worked as expected. A comparison was put up between the implementation of
FIR and IIR filters bank designs which concluded in very similar output results.
Figure 3. Depicts the filterbank inside the
subsystem..
Figure 4. Shows the output signal
using the spectrum scope.
Project summary
This thesis is targeted at designing and implementing
a software electronic model of the human ear.
Initially, the human ear is studied along with all its
functions. Two effective computational ear models,
Lyon’s and Patterson’s are investigated while Lyon’s
model is chosen to design the ear model. The
software simulations are done using Simulink. Since
the human ear has the audible range of 20Hz to
20kHz, the total frequency range in the model is also
the same. The segregation of the frequency bands
among the filters are divided according to the
sensitivity of the human ear.
Project Objectives
• Study of the human auditory system which is the
root of this research.
• Investigate the existing achieved auditory
research and learn the Lyon’s auditory model.
• Design the cochlea gram based on cascade only
Lyon’s cochlear model on Simulink.
• Validation of various components of the design
under different conditions such as:
‣ Response to human speech: Human ear has better
sensitivity to speech when compared to other
frequencies. This test will help in test it.
‣ Response to the entire hearing spectrum
‣ Response to frequencies outside the hearing
spectrum
Project Conclusion
The anatomy of human ear was analyzed. Various
filter designs such as FIR, IIR and their
implementations were studied by simulating on
Simulink and understanding the responses. The
spread factor was derived such that the response of
one filter does not affect the response of other filters
because all of them are mutually coupled. The
completed design was then subjected a number of
tests across all the corners of the input spectrum. The
results obtained matched the understanding of the
Lyon’s cochlea model.