Voice conversion based on continuous frequency warping and magnitude scaling