Abstract: A voice conversion scheme using isolated speaker model without need of joint training and parallel speech corpus is proposed.In system training,Structured Gaussian Mixture Model is trained for every speaker using speaker dependent speech samples.In conversion,the Gaussian components of source and target speakers models are first aligned by a technology named as Acoustic Universal Structure,then transform formula is achieved.Experiments show the proposed method not only has the equivalent conversion performance,but also fewer computational consume and more flexible system extension ability.