A Novel Linear-Polynomial Kernel to Construct Support Vector Machines for Speech Recognition

Abstract

Problem statement: To accept the inputs as spoken word utterances uttered by various speakers, recognize the corresponding spoken words and initiate action pertaining to that word. Approach: A novel Linear-Polynomial (LP) Kernel function was used to construct support vector machines to classify the spoken word utterances. The support vector machines were constructed using various kernel functions. The use of well known one-versus-one approach considered with voting algorithm. Results: The empirical results compared by implementing various kernel functions such as linear kernel function, polynomial kernel function and LP kernel functions to construct different SVMs. Conclusion: The generalization performances based on the One-versus- One approach for speech recognition were compared with the novel LP kernel function. The SVMs using LP kernel function classifies the spoken utterances very efficiently as compared to other kernel functions. The performance of the novel LP kernel function was outstanding as compared to other kernel functions.