SINGAN: Singing Voice Conversion with Generative Adversarial Networks

Singing voice conversion (SVC) is a task to convert the source singer's voice to sound like that of the target singer, without changing the lyrical content. So far, most of the voice conversion studies mainly focus only on the speech voice conversion that is different from singing voice conversion. We note that singing conveys both lexical and emotional information through words and tones. It is one of the most expressive components in music and a means of entertainment as well as self expression. In this paper, we propose a novel singing voice conversion framework, that is based on Generative Adversarial Networks (GANs). The proposed CAN-based conversion framework, that we call SINGAN, consists of two neural networks: a discriminator to distinguish natural and converted singing voice, and a generator to deceive the discriminator. With CAN, we minimize the differences of the distributions between the original target parameters and the generated singing parameters. To our best knowledge, this is the first framework that uses generative adversarial networks for singing voice conversion. In experiments, we show that the proposed method effectively converts singing voices and outperforms the baseline approach.