Our objective is to design a close-set system that would recognize five designated speakers by using some of the DSP techniques we learned in MATLAB. To do this, we have a pre-recorded template that stores each speaker’s distinctive features. We will use this template to do a mix and match with the speakers in our system.
Speaker identification refers to the process of identifying an individual by extracting and processing information from his speech. This is a fascinating area of research. From speech itself, we can deduce quite accurately whether the speaker is male or female, adult or child. It is also possible to detect the emotional state, and attitude of the speaker. History of this specific area is not quite well known because not many people succeeded in developing this system. Our motivations came from our interest for DSP in the field of Electrical Engineering. To be a little more specific, the motivation came from the courses that we have learned in signal processing at UC Riverside. Our goals are developed as we go along because we wish to accomplish quite a few things with the speaker identification system. However, our primary goal is finish the project with the satisfaction of accomplishing something that was at a level higher than what we have expected. The resulting performance is based upon several main factors: how well we train our Artificial Neural Network, how representative the speech features are, and how consistent the data are. The ANN determines how accurate and precise our comparison will be with our stored template. If the ANN finds the speaker in the template, we’ll be able to display his name on the monitor. We will show our approach with step by step explanations in our design solution. We will also have comments implemented in our code to explain certain significant sections. The steps we have taken will show the reader what we have accomplished. We hope that maybe one day, people will expand on what we have...
Please join StudyMode to read the full document