Text this: A Stochastic Residual-Based Dominant Component Analysis for Speech Enhancement