Perceive and predict: self-supervised speech representation based loss functions for speech enhancement