الفهرس Only 14 pages are availabe for public view
 |  | from | 75 |  |  |
| | | from | 75 | | |
Abstract
Human language carries various kinds of information. In human telephone interaction, the detection of the importance of the message left on the answering machine is crucial for determining the priority of the messages and hence arranging the messages according to their priorities. Three main categories of messages are spotted; urgent messages which have the higher priorities (i.e. emergency calls), normal messages which contains the commonly and daily used words, and not urgent messages which have the lowest priorities.
The problem is how to build a system able to identify the mood of the person leaving a message on the answer machine, as the detection of the emotional state of a speaker leaving a voice message on the answer machine as reflected in his or her utterances is needed to identify the priority of the message, robust system that deals with the speaker variability due to variable speaker’s moods which is considered one of the biggest problems in automatic speech recognition is needed. Also, the system has to deal with the speech contained in the message, recognize it or in other words, understand it and use the understood speech to arrange the messages.
This thesis introduces a proposed design of the system that comprises two subsystems; the first subsystem deals with the nature of the utterance and the voice of the person using Adaptive Neuro-Fuzzy Inference mood identifier. The second subsystem is utilized to recognize the speech contained in the message. This second subsystem deals with the speech of the message and based on hybrid ANN/HMM word spotter with a phoneme HMM network for word verification, which uses HMM phoneme network-to-MLP neural network probability estimator. The thesis analysis is based on Arabic sentences with different male and female speakers.