ZSE-VITS: A Zero-Shot Expressive Voice Cloning Method Based on VITS

Voice cloning aims to synthesize the voice with a new speaker’s timbre from small amount of speech. Current methods, which focus on modeling speaker timbre, can speech similar timbres. However, prosody these methods is flat, lacking expressiveness and ability control cloned To solve this problem, we propose novel method ZSE-VITS (zero-shot expressive VITS) based end-to-end synthesis model VITS. Specifically, use VITS as backbone network add recognition TitaNet encoder realize zero-shot cloning. We explicit information avoid effects adjust using prediction fusion directly. widen pitch distribution train datasets augmentation improve generalization model, fine-tune predictor alone in emotion corpus learn various styles. The objective subjective evaluations open show that our generate more artificially without affecting similarity timbre.