We propose MediumVC, an utterance-level method towards A2A. Before that, we propose SingleVC to perform A2O tasks(Xi → Ŷi) , Xi means utterance i spoken by X). The Ŷi are considered as SSIF. To build SingleVC, we employ a novel data augment strategy: pitch-shifted and duration-remained(PSDR) to produce paired asymmetrical training data. Then, based on pre-trained SingleVC, MediumVC performs an asymmetrical reconstruction task(Ŷi → X̂i). Due to the asymmetrical reconstruction mode, MediumVC achieves more efficient feature decoupling and fusion. Experiments demonstrate MediumVC performs strong robustness for unseen speakers across multiple public datasets. The more details can access paper.
This page provides converted speech samples. Contrast methods include MwuS (MediumVC with untrained SingleVC), MwoS (MediumVC without SingleVC), FragmentVC, AutoVC and AdaIn-VC.
VCTK
F2F, The Norsemen considered the rainbow as a bridge over which the gods passed from earth to their home in the sky.
F2M, We also need a small plastic snake and a big toy frog for the kids.
M2F, Since then physicists have found that it is not reflection, but refraction by the raindrops which causes the rainbows.
M2M, Many complicated ideas about the rainbow have been formed.
Source
Target
MediumVC
MwuS
MwoS
FragmentVC
AutoVC
AdaIn-VC
LibriSpeech
F2F, A voice on the beach under the cliff began to sing.
F2M, The two started to walk back along the road toward town.
M2F, This was impregnable, and admitted of neither objection nor restriction.
M2M, From a cupboard he pulled out one of his old shirts, which he tore in pieces.
Source
Target
MediumVC
MwuS
MwoS
FragmentVC
AutoVC
AdaIn-VC
VCC2020
F2F, Moroccan agriculture enjoys special treatment when exporting to Europe.
F2M, How does the thing cut the true wall?
M2F, Small investors will also be affected, traders said.
M2M, Modern electronics has become highly dependent on inorganic chemistry.