Minimization of speech alignment error by iterative transformation for speaker adaptation