An Optimum Database for Isolated Word in Speech Recognition System

Abstract

Speech recognition system (ASR) is a technology that allows computers receive the input using the spoken words. This technology requires sample words in the pattern matching process that is stored in the database. There is no reference as the fundamental theory to develop database in ASR. So, the research of database development to optimize the performance of the system is required.  Mel-scale frequency cepstral coefficients (MFCCs) is used to extract the characteristics of speech signal and backpropagation neural network in quantized vector is used to evaluate likelihood the maximum log values to the nearest pattern in the database.  The results shows the robustness of ASR is optimum using 140 samples of data reference for each word with an average of accuracy is 99.95% and duration process is 27.4 msec.  The investigation also reported the gender doesn’t have significantly influence to the accuracy.  From these results it concluded that the performance of ASR can be increased by optimizing the database.