Describir: All’s Well That FID’s Well? Result Quality and Metric Scores in GAN Models for Lip-Synchronization Tasks