Efficient Adversarial Audio Synthesis via Progressive Upsampling