EMOTIONAL VOICE CONVERSION WITH WAVELET TRANSFORM USING DUAL SUPERVISED ADVERSARIAL NETWORKS

摘要

This paper proposed an emotional VC methodrnusing dual supervised adversarial networks (dual-rnSANs) with MCC and CWT-F0 features. In orderrnto obtain a better training results, we process 2DrnMCC features and 2D AS-CWT features, and sep-rnarately train them with the proposed dual-SANs.rnThe results shows that the dual supervised learningrnhas a great affect on the dual emotional VC tasks.