Speech recognition is very popular field of research and speech classification improves the performance
for speech recognition. Different patterns are identified using various characteristics or features of speech
to do there classification. Typical speech features set consist of many parameters like standard deviation,
magnitude, zero crossing representing speech signal. By considering all these parameters, system
computation load and time will increase a lot, so there is need to minimize these parameters by selecting
important features. Feature selection aims to get an optimal subset of features from given space, leading to
high classification performance. Thus feature selection methods should derive features that should reduce
the amount of data used for classification. High recognition accuracy is in demand for speech recognition
system. In this paper Zernike moments of speech signal are extracted and used as features of speech signal.
Zernike moments are the shape descriptor generally used to describe the shape of region. To extract
Zernike moments, one dimensional audio signal is converted into two dimensional image file. Then various
feature selection and ranking algorithms like t-Test, Chi Square, Fisher Score, ReliefF, Gini Index and
Information Gain are used to select important feature of speech signal. Performances of the algorithms are
evaluated using accuracy of classifier. Support Vector Machine (SVM) is used as the learning algorithm of
classifier and it is observed that accuracy is improved a lot after removing unwanted features.