A multiscale invariant method for speaker independent speech recognition in telecom environment