Towards an Interpretable Representation of Speaker Identity via Perceptual Voice Qualities