Using joint training speaker encoder with consistency loss to achieve cross-lingual voice conversion and expressive voice conversion - View it on GitHub