Encoding Musical Style with Transformer Autoencoders