Kostopoulos, George

The primary objective of this thesis is to design and construct a program which will permit the user to obtain immediate results for a microcellular communication traffic engineering problem, although he may not have a thorough knowledge of telephone environment. The program is related to multi-channel, trunked communication systems, and it associates the number of channels and users as a function of system grade of service by implementing Poisson, Erlang B, Erland C and Crommelin-Pollaczek traffic models. The user will be able to choose between the different traffic models mentioned above via a menu. Then he will be asked to enter the parameters for the unknown variable(s) he wants to find. At this point, he will have to enter some other parameter values for which he is designing, and finally, the computer program will give the desired results (i.e., the number of servers needed, possible total traffic offered to the system, probability of loss or delay).

An isolated word recognition system for the words zer o
through nine has been implemented using an IBM Personal
Computer AT with a Data Translation DT2821 12-bit data
acquisition card . The software has been written f o r the
80286 based machine in assembly and compiled basic. Since
this research was aimed at defining techniques that could
work for speech recognition, real time recognition was not
a goal.
Three techniques were tested for speech recognition.
The first method tried to match entire words, using the
fast Fourier transform, the fast Walsh transform, the
average magnitude function, and the zero crossing rate.
The second method used the short-time fast Fourier
transform to recognize the phonemes, then the words. The
third method used a digital filter generated from I in ear
predictive coding coefficients and the zero crossing rate
to recognize the phonemes and words. A recognition rate of
98 percent was achieved for a single user for the words
zero through nine.

In speech analysis, a Voiced-Unvoiced-Silence (V/UV/S) decision is performed through pattern recognition, based on measurements made on the signal. The examined speech segment is assigned to a particular class, V/UV/S, based on a minimum probability-of-error decision rule which is obtained under the assumption that the measured parameters are distributed according to a multidimensional Gaussian probability density function. The means and covariances for the Gaussian distribution are determined from manually classified speech data included in a training set. If the recording conditions vary considerably, a new set of training data is required. With the assumption that all three classes exist in the incoming speech signal, this research describes an automatic parametric learning method. Such a method estimates the means and covariances from the incoming speech signal and provides a reliable classification in any reasonable acoustic environment. This approach eliminates the necessity for the manual classification of training data and has the capability of being self-adapting to the background acoustic environment as well as to speech level variations. Thus the presented approach can be readily applied to on-line continuous speech classification without prior recognition.