PUB - Publikationen an der Universität Bielefeld

Disentangling speaker and content attributes of a speech sig- nal into separate latent representations followed by decoding the content with an exchanged speaker representation is a pop- ular approach for voice conversion, which can be trained with non-parallel and unlabeled speech data. However, previous ap- proaches perform disentanglement only implicitly via some sort of information bottleneck or normalization, where it is usually hard to find a good trade-off between voice conversion and con- tent reconstruction. Further, previous works usually do not con- sider an adaptation of the speaking rate to the target speaker or they put some major restrictions to the data or use case. There- fore, the contribution of this work is two-fold. First, we employ an explicit and fully unsupervised disentanglement approach, which has previously only been used for representation learn- ing, and show that it allows to obtain both superior voice conver- sion and content reconstruction. Second, we investigate simple and generic approaches to linearly adapt the length of a speech signal, and hence the speaking rate, to a target speaker and show that the proposed adaptation allows to increase the speaking rate similarity with respect to the target speaker.