期刊名称：International Journal of Computer Trends and Technology
电子版ISSN：2231-2803
出版年度：2013
卷号：4
期号：9-1
出版社：Seventh Sense Research Group
摘要：Kernels modelbased is an advanced technique for recognizing speech for large amount of training data in continuous speech recognition. This research describes a particular model in this framework, kernel based structured support vector machine (SSVM).This shows that how it can be applied to medium to large vocabulary recognition tasks. Here a context –dependent models and high dimensional feature of derivative kernels are used .This extends the previous work with combined generative and discriminative classifiers .The derivative kernel feature is extracted. And these extracted features can be segmented by using Viterbilike scheme. Viterbilike scheme is described for obtaining optimal segmentation. Finally a training algorithm can be incorporated into large margin criterion. The performance of SSVM is evaluated to aurora 2 speech recognition task. The experimental result provides kernel based SSVM provides best performance result over SSVM.
关键词：StructuredSVM; kernel methods; Log Linear Models; AURORA 2
