Transfer Learning Approaches For Feature Denoising And Low-Resource Speech Recognition