Dissertation title: "Speech recognitionProcessing Dialog System for Database access though Telephone"

Author: Ioannis G. KARRAS
e-mail: ikarras@yahoo.com
Introduction
This speaker independent recognising system application has been designed for fast speech recognition and high recognition accuracy using a small vocabulary up to 64 words in Greek language. The system has been evaluated with test results. These results have been obtained by special developed training and correcting methods based on statistics taken by experiment observation and benchmarks. One PC has been used to repeat words with the same characteristics as velocity animation, telephone microphone distance, word variability etc for repeatable tests on real time to prove the real time Word Correct Error. These tests are not possible in real time otherwise because human been can not repeat just the same word many times in the same way. After the optimisation of the parameter values in configuration files, a series of programs and recording message files have been created. These are controlled by other scripts, which run in the background to assure the continuity of the program flow. Other scripts are responsible for the interactive and communication features of the system. In this dissertation is not only built a recognition system from scratch but are also investigated faster techniques that include approximations that increase the recognition accuracy of the system, in fast recognition speed mode, prior any further training action.

 
Go back to main page