Perceptual learning of synthetic speech