Many-to-Many Voice Transformer Network

This paper proposes a voice conversion (VC) method based on sequence-to-sequence (S2S) learning framework, which enables simultaneous of the characteristics, pitch contour, and duration input speech. We previously proposed an S2S-based VC using transformer network architecture called (VTN). The original VTN was designed to learn only mapping speech feature sequences from one speaker another. Here, main idea we propose is extension that can simultaneously mappings among multiple speakers. extension, many-to-many VTN, us fully use available training data collected speakers by capturing common latent features be shared across different It also allows introduce loss identity ensure sequence will remain unchanged when source target indices are same. Using this particular for model has been found extremely effective in improving performance at test time. conducted experiments our obtained higher sound quality similarity than baseline methods. model, with slight modification its architecture, handle any-to-many tasks reasonably well.