Fig. 5From: Multimodal voice conversion based on non-negative matrix factorizationLip imagesBack to article page