Evaluation on data — Speaker dependability approaches for speech recognition tasks

In this feasibility study, four approaches in implementing speech recognition system for speech modeling are proposed. These approaches are based on data — speaker dependability of speech recognition system. Desired information or features in Mel Frequency Cepstral Coefficient (MFCC) are extracted from the speech samples. For pattern matching, Support Vector Machine (SVM) classifier is used to perform the speech patterns classification. The main objective of this study is to evaluate the performance of data- speaker dependability approaches in term of percentage of accuracy and Mean Squared Error (MSE). Result suggests that data dependent — speaker dependent and data independent — speaker dependent approaches are more suitable to be used in speech recognition process as they both gave an accuracy of >95% and MSE <0.2.