Classification of multidimensional trajectories for acoustic modeling using support vector machines

In this paper, we address the issues in classification of varying duration segments of context dependent subword units of speech using support vector machines. Commonly used methods for mapping the varying duration segments into fixed dimension patterns may lead to loss of crucial information necessary for classification. We propose two methods in which the segment of a subword unit of speech is considered as a trajectory in a multidimensional space. In the first method. a pseudo-innerproduct between two trajectories in the Mercer kernel feature space is used as the kernel operation in construction of support vector machines for classification. In the second method, a fixed dimension pattern vector derived from an outerproduct operation on a trajectory is given as input to the support vector machines. The performance of the proposed methods is studied on recognition of a large number of consonant-vowel units of speech in a broadcast news corpus.