<p>Sensorineural hearing loss imparts two serious hearing deficits in hearing-impaired
people: reduced dynamic range of hearing and reduced frequency selectivity. Psychophysically,
these deficits render loss of speech audibility and speech intelligibility
to a hearing-impaired person. Studies of an impaired cochlea in cats have shown that
the hearing loss originates from damage to or complete loss of inner and outer hair
cells. Neurophysiology of an impaired cochlea in cats shows that the tuning curves
of the auditory nerve fibers become elevated and broadened. Amplification in hearing
aids has been used to restore audibility in hearing-impaired people. Multiband
compression has been commercially available in conventional hearing aids to compensate
for the reduced dynamic range of hearing. However, little has been achieved to
improve the intelligibility of speech in the hearing-impaired people. The aim of this
thesis is to restore not only the speech audibility in a hearing-impaired person, but
also to improve their speech intelligibility through some hearing-aid signal processing.
The compensation technique used in this thesis for speech intelligibility is based on
the hypothesis that a narrowband response of the auditory nerve fibers to speech
signals ensure phonemic discriminability in the central nervous system.</p><p>Miller et al. [1999] have proposed contrast-enhancing frequency shaping ( CEFS) to
compensate for the broadband responses of the fibers to first and second formants (Fl
and F2) of a speech stimulus. Bruce [2004] has shown that the multiband compression
can be combined with CEFS without counteracting each other. In Bruce's algorithm,
a multiband compressor is serially combined with a time-domain CEFS filter. The
MICEFS algorithm, herein presented, is a combination of multiband compression and
an improved version of CEFS implemented in the frequency domain. The frequency
domain implementation of MICEFS has improved the time delay response of the
algorithm by 10 ms as compared to series implementation proposed by Bruce. The
total time delay of the MICEFS algorithm is 16 ms, which is still longer than the standard time delay of 10 ms in hearing aids. The MICEFS algorithm was tested on
a computational model of auditory periphery [Bruce et al., 2003] using a synthetic
vowel and a synthetic sentence. The testing paradigm consisted of five conditions:
1) unmodified speech presented to a normal cochlea; 2) speech modified with halfgain
rule presented to an impaired cochlea; 3) CEFS modified speech presented to
the impaired cochlea; 4) speech modified with MICEFS presented to the impaired
cochlea, and; 5) MICEFS-modified speech with some added noise in the formant
estimation presented to an impaired cochlea. The spectral enhancement filter used in
MICEFS has improved the synchrony capture of the fibers to the first three formants
of a speech stimulus. MICEFS has also restored the correct tonotopic representation
in the average discharge rate of the fibers at the first three formants of the speech.</p> / Thesis / Master of Applied Science (MASc)
Identifer | oai:union.ndltd.org:mcmaster.ca/oai:macsphere.mcmaster.ca:11375/21878 |
Date | 07 1900 |
Creators | Ansari, Shahabuddin |
Contributors | Bruce, Ian C., Electrical and Computer Engineering |
Source Sets | McMaster University |
Language | en_US |
Detected Language | English |
Type | Thesis |
Page generated in 0.0016 seconds