Applying Sparse KPCA for Feature Extraction in Speech Recognition