Recognition of Speech Isolated Words Based On Pyramid Phonetic Bag of Words Model Display and Kernel-Based Support Vector Machine Classifier Model

This study aimed to improve the classification of individual (isolated) words, and specifically, the numbers from one to twenty. In this study, a strong model was suggested to gain a unified view of voice. It is based on the idea of phonetic bag for voice that has been developed into a pyramid state. The pyramid idea can model temporal relationships. One of the problems of Support Vector Machine to classify words is its inability to model temporal relationships unlike hidden Markov models. Using the BOW-based pyramid idea in the extraction of the display containing temporal information of voice, the SVM can be given the capability of considering the time relationships of speech frames. One of the main advantages of Support Vector Machine model is its fewer parameters than the hidden Markov model. As the experiments results have shown, it has much higher accuracy than the hidden Markov model in applications such as the recognition of single words, where the data set volume is limited. Using the pyramid BOW idea, the accuracy of SVM-based method can be increased as 20% compared to previous methods.

کلیدواژه ها:

Speech recognition ، Isolated words recognition ، Classification of speechIntroduction ، Display of phonetic bag of words ، Support vector machine