A COMPARISION OF VARIOUS NOISE REDUCTION

            METHODS IN MOBILE COMMUNICATION

                    Tilak Nissanka

In almost all practical situations, the received speech
waveform contains some form of noise component. The noise
may be a result of the finite precision involved in
coding the transmitted waveform, or due to the addition
of acoustically coupled background noise. Depending on
the amount and type of noise, the quality of the received
waveform can range from being slightly degraded to being
annoying to listen to, and finally to being totally
unintelligible.
 
Of late the problem of removing the unwanted noise
component from a received signal has captured the
imagination of many scholars. Enhancement of speech degraded
by acoustic noise has been the topic of extensive research
through the late 1970s up to today with varying amount of
success. In this thesis project some of the techniques that
have been proposed have been analysed with respect to their
inherited merits and demerits when subjected to various types
of additive background noise. Attempts have also been made to
increase the intelligibility of the noisy speech signal by
combining the ideas of different noise-reduction techniques.
 
Noise reduction methods analysed in this paper includes
spectral subtraction technique based on short-time spectral
amplitude estimation, adaptive noise filtering using a RLS
filter, wavelet techniques based on threshold method and
spectral subtraction as well as an attempt to further enhance
outputs from wavelet techniques (these outputs were already
available for us from a previous thesis) by means of weighting
them optimally. The background noise (both white and coloured
noise have been used) is digitally added to the speech.
 
Results indicate that while spectral subtraction is capable of
removing considerable amount of background noise, it also
introduces an undesirable distortion to the processed speech
known as ``the musical noise'', due to random variations of the
noise spectrum. An assessment of wavelet techniques is not
within the scope of this paper although it is known that
threshold method's applications are largely restricted to speech
signals degraded by white noise. However, the use of spectral
subtraction in the wavelet domain has shown promising results
when applied to signals contaminated by coloured noise. Results
from adaptive noise filtering have largely been dissatisfactory,
due to the fact that noise components of wavelet outputs are not
entirely uncorrelated as required by adaptive noise cancelling
technique. Results obtained through weighting of wavelet outputs
are, however, encouraging, at least if our intention is to
improve the SNR value.