Masked Audio Generation using a Single Non-Autoregressive Transformer