Linear discriminant analysis, principipal component analysis and sequential forward search for speaker feature extraction: a comparative study