Model Extraction Attack against Self-supervised Speech Models