Using multiple reference audios and style embedding constraints for speech synthesis

Re-assign community