A large self-supervised speech representation model from Microsoft for denoising and speaker-aware tasks.