Voice Recognition Improvement Using Kaiser Windowing Method

Ho, Woei Loong (2018) Voice Recognition Improvement Using Kaiser Windowing Method. Final Year Project (Bachelor), Tunku Abdul Rahman University College.

Text HO WOEI LOONG - FULLL THESIS.pdf Restricted to Registered users only Download (2MB)

Abstract

The concept of voice recognition was establishing human-machine voice communication. Research on voice recognition was always focused on how to transfer speech into a form that can be read by a machine. Algorithms, probability models and neural networks were researched and developed on to help the computer understand what to do with all the sample data for it to determine the speech being spoken to them. With all these contributions that have aided in the advancement of voice recognition technology, the reliability of these systems are still not practical enough to be put to use in an emergency situation. On the other hand, input speech signals can be enhanced to create a better input sample before using it for signal processing feature extraction. Signals are normally processed frame-by-frame by windowing function to smoothen the signal. Using a window that is able to change, Kaiser window, its area of frequency coverage to optimised for an input signal can affect the performance of voice recognition system. A normal window, which is a hamming window, is compared with the Kaiser window to show the difference of performance of a voice recognition system. The change in accuracy of a voice recognition system can be observed, along with the effects, advantages and drawbacks of using both of these windowing algorithms.