StarGAN-VC: Non-parallel many-to-many voice conversion with star generative adversarial networks
This addresses voice conversion for applications like speech synthesis, offering a more efficient and flexible approach, though it builds incrementally on existing GAN techniques.
The paper tackled non-parallel many-to-many voice conversion by proposing StarGAN-VC, a method using StarGAN that requires no parallel data and learns mappings across domains with a single generator, achieving higher sound quality and speaker similarity than a state-of-the-art method in subjective evaluations.
This paper proposes a method that allows non-parallel many-to-many voice conversion (VC) by using a variant of a generative adversarial network (GAN) called StarGAN. Our method, which we call StarGAN-VC, is noteworthy in that it (1) requires no parallel utterances, transcriptions, or time alignment procedures for speech generator training, (2) simultaneously learns many-to-many mappings across different attribute domains using a single generator network, (3) is able to generate converted speech signals quickly enough to allow real-time implementations and (4) requires only several minutes of training examples to generate reasonably realistic-sounding speech. Subjective evaluation experiments on a non-parallel many-to-many speaker identity conversion task revealed that the proposed method obtained higher sound quality and speaker similarity than a state-of-the-art method based on variational autoencoding GANs.