This is the demo page for the NANSY++ unofficial implementation of the open-source repository published by MWM.
Following section showcases zero-shot voice conversion ability of backbone checkpoints trained on HifiTTS with the open-source repo (”OS repo”) and another trained on internal data (”our best”). The inferencer class was used to synthesize results. Source and target audio samples are unseen examples from VCTK corpus that can be either extracted from full dataset or found in the static sub-directory.
Source
Target
NANSY++ (OS repo)
NANSY++ (our best)
Speaker
p238
p248
p261
p326
p347
Reference Input
GT
NANSY-TTS w/ NANSY++(vctk)
NANSY++ (OS repo)