Firstly in this part of my project I took noise in the form of a
sinusoidal waveform since mostly noise is supposed to be some form of
oscillatory nature if it is coming from vibrations and this clearly
matches the problem that I have chosen. In the first case I fixed the
maximum amplitude of my noise to be 0.1. The 
was fixed at
=0.065/(length*samp*samp), where samp is the maximum amplitude
of my signal. After doing the simulations I heard first the signal,
then the noise that I had introduced, thirdly the signal and the noise
combined and finally the output of my filter. My filter did a good job
by recognising the original signal almost perfectly. When I plotted
the error versus the k value, the maximum amplitude of my error was
about 0.02 (refering to figure-). The signal power is proportional
to amplitude squared. The maximum amplitude of my speech signal is 0.1483.
Dividing 0.1483 by 0.02 gives about 7.42 and the amplitude squared of
that gives approximately 55, which implies that the noise power is
far below now. All the plots are appended.
Now next I increased the maximum noise amplitude to 0.2, which is higher
than the original speech signal. I again checked if my filter was
working or not by listening to the sounds of the original signal, noise
signal, signal + noise signal and finally the output signal. It still
worked perfectly well. The plots for the original speech signal, noise
signal, signal+noise signal, the output signal and the error signals
are shown. The is taken to be the same and the length of the
filter was taken to be 6.
gsm@ee.duke.edu
Fri Mar 08 18:41:27 1996